CN provider accounts (Kimi/Zhipu/DeepSeek) and the monitor quota view each
hand-rolled their own usage bar, so the same account column rendered three
different label styles, bar widths and threshold palettes. Both now reuse
UsageProgressBar, the component the other five platforms already share.
Also move the default utilization thresholds from >=100 red / >=80 amber to
>=90 / >=75. Turning red only once the quota is exhausted is an after-the-fact
notice rather than a warning. The remainingCapacity mode keeps its 20/50 pair,
which is already a warning semantic.
- UsageProgressBar: add labelWidth 'fixed' | 'auto' so the monitor view can
render composite labels ("Pro/7d") without breaking column alignment; the
fixed branch is byte-identical, so the existing platforms are untouched
- drop the superseded utilizationColor/utilizationTextColor/formatReset copies
- shorten CN window labels to 5h / 7d and remove the now-dead resetSoon keys
The cache token breakdown tooltip in `UsageStatsCards` was hidden with
`opacity-0`. An element with zero opacity is invisible but still takes part
in layout, so its fixed `w-56` (224px) width kept contributing to the
document's scrollable overflow.
The tooltip is anchored with `left-1/2 -translate-x-1/2`, and on mobile the
stats grid falls back to `grid-cols-2`, which puts the trigger icon close to
the right edge of the viewport. Centering a 224px box on that point pushes it
roughly 26-49px past the right edge, so `/usage` and `/admin/usage` gained a
phantom horizontal scroll with nothing visible to explain it.
Switching the hidden state to `hidden` (`display: none`) takes the tooltip out
of the flow entirely, while `group-hover:block` / `group-focus:block` keep the
existing hover and keyboard-focus behaviour. The 150ms opacity transition is
dropped, since `display` cannot be transitioned.
Verified over Chrome DevTools Protocol at 360/390/414px: `documentElement.scrollWidth`
goes from 424 back to 390 (equal to `clientWidth`) with no other overflowing
elements remaining, and the tooltip still renders at 224x86 on focus. A sweep of
all 39 application routes confirmed these were the only two affected pages.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
grok-build CLI recap-style side-calls (turn-summary / title-refresh) send
a fresh per-call X-Grok-Conv-Id label but set body prompt_cache_key to
the parent session id, so a direct xAI connection shares the main turn's
server-side prompt cache prefix.
sub2api's explicitGrokCacheSeed preferred X-Grok-Conv-Id over the body
key, which fragmented the cache identity on every side-call and forced a
full-price replay of the whole conversation (~300K+ tokens each, ~2.3%
cache hit rate observed over 44 side-calls / 4.9M input tokens in 7
days — about a third of all grok-4.6 input).
Priority is now: CC session > generic session headers > body
prompt_cache_key > X-Grok-Conv-Id > explicitKey > previous_response_id.
IDE header-stability tests updated; new regression test locks in the
side-call/parent key-sharing contract.
Team GLM Coding Plan accounts fail the quota probe with
"当前用户不存在coding plan" because the personal-plan query path
does not carry team context.
- quota probe: when credentials contain zhipu_organization, switch the
Zhipu quota endpoint to the team variant (?type=2 + bigmodel-organization
/ bigmodel-project headers); personal plans keep the existing path
- account create/edit modals: optional organization/project ID inputs for
zhipu coding-plan accounts, persisted in credentials; edit modal
backfills them and clears back to personal plan when emptied
- add a click-to-open tooltip tutorial (zh/en) explaining how to grab the
IDs from the bigmodel.cn web console via the /api/biz/v1/organization
network request, with a sample URL mapping org-/proj_ segments to fields
The passthrough streaming loop buffers leading events in `pendingLines` until
the first visible output, so a client receives **zero bytes** — not even the
HTTP response headers, since gin's ResponseWriter only commits them on the
first write — for as long as the upstream takes to produce its first visible
event.
The Forward path already recognises this exact hazard and solves it with a
downstream keepalive:
// Track downstream writes separately from upstream reads: pre-output
// failover can buffer response.created / response.in_progress, so
// keepalive must be based on downstream idle time.
The passthrough path has the same buffering but never starts a keepalive.
Reasoning models routinely think for several minutes before emitting a visible
event, so intermediate proxies close the connection on idle timeout.
Production evidence from one deployment (12h window, gpt-5.6 family):
* 44 `/v1/responses` requests produced their first visible output only after
600–900 s, each carrying 30k–50k output tokens — the upstream had finished
the work.
* Every one of them was killed by an intermediate nginx `proxy_read_timeout`
(600 s) and returned 504. The user received nothing.
* Cutting early does not even save resources: the code deliberately keeps
draining the upstream for usage after the client disconnects.
Fix: start the existing keepalive helper for the passthrough stream as well.
* `StartOpenAICompactSSEKeepalive` keeps its compact-marker gate; the gate-free
body is extracted as `startOpenAISSEKeepalive` for callers that already know
they are in an SSE context.
* The passthrough loop starts it when `gateway.stream_keepalive_interval > 0`
and stops it immediately before its first real write, so the ResponseWriter
is never shared.
Pre-output failover is unaffected: keepalive bytes are already excluded by
`OpenAICompactKeepaliveAdjustedWrittenSize`, which `openAIStreamClientOutputStarted`
on this very path already consults (#3887). No pendingLines and no
account-specific headers are ever emitted by a keepalive.
Behaviour is unchanged when `gateway.stream_keepalive_interval` is 0.
Public groups have always been bindable by every user: CanBindGroup returned
true for any non-exclusive group, and user_allowed_groups only ever carried the
exclusive groups an admin had granted. Admins had no way to hand a single user
a subset of the public groups short of converting a group to exclusive, which
changes it for everyone already using it.
A user now carries restrict_public_groups. While it is false, which is the
default and what every existing row migrates to, nothing changes: every public
group stays bindable. Once an admin turns it on for a user, that user's public
groups are narrowed to the ones listed in user_allowed_groups, the same table
that already gates exclusive groups.
The flag is an administrative control, so it rides on the admin user DTO only
and leaves the shape of the end-user endpoints alone.
The model plaza filter honours the flag as well, so a restricted user is not
shown groups they would be refused when binding a key. Anonymous visitors have
no user record and keep the previous view.
Enforcement rides on the existing CanBindGroup choke point, so it covers key
creation, key updates, and per-request authorization together. An API key bound
to a group that is later withdrawn stops working at request time rather than
lingering as a key that can be listed but not used.
The admin dialog gains a toggle over the public group list. Turning it off
re-checks every public group, so an admin cannot save a list that reads as
restrictive while the restriction itself is disabled.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
When running a custom build or pre-release version (e.g. v0.1.183-custom),
parseVersion failed to parse the patch segment because of the non-numeric suffix,
causing compareVersions to incorrectly report an available update.
This commit strips any hyphenated suffix before splitting SemVer components.
Signed-off-by: WUJI-Labs <zoujincool@gmail.com>
Merge origin/main into contrib/routed-codex-model-catalog.
Single conflict in openai_gateway_forward.go: main added
ClearActualOpenAIUpstreamEndpoint + SetActualOpenAIUpstreamEndpoint
(commit 4795650d2) at the same insertion point where the PR added
filterOpenAIResponsesNoneReasoningEffortForAccount. Both changes
are independent — endpoint observability first, then body filtering.
Copy cached API-key manifests before group-specific mutation, use DeepSeek
model IDs for Codex fallbacks, omit unsupported config.toml effort, and
drop wildcard mapping keys from generated catalogs.
Co-authored-by: Cursor <cursoragent@cursor.com>
Keep catalog membership on schedulable accounts and intersect advertised
capabilities across all active group members that map an alias. Preserve
upstream plugin, service-tier, and models-list body-limit changes.
Co-authored-by: Cursor <cursoragent@cursor.com>
Intersect advertised alias capabilities across all active group members
that map the model, not only currently schedulable accounts, so rate-limits
cannot widen image input or context windows.
Co-authored-by: Cursor <cursoragent@cursor.com>
Users only see the reasoning effort they requested. Admins still see
the requested value plus the forwarded mapped value, matching the
model column's requested vs upstream split.
The raw Chat Completions passthrough returned a nil error whenever the upstream
body ended, so an upstream that cut the SSE stream mid-generation surfaced as
HTTP 200 with usage 0/0: the client kept a half-finished answer, the request was
not counted as an SLA failure, and ops saw nothing. Observed against
opencode.ai/zen/go (deepseek-v4-pro): 11 of 1346 streams in 24h recorded a first
token but no final usage, alongside a concurrent `connection reset by peer`
against the upstream edge.
Track whether the upstream ever emitted a terminal signal — [DONE], a usage
chunk, or a finish_reason — and treat a stream that ends without one as an
upstream failure. Matching only [DONE] would misfire on compatible upstreams
that EOF right after their last chunk and would drop billing for tokens that
were actually delivered, so all three count.
- nothing written to the client yet: UpstreamFailoverError, retried on another
account with no client-visible half stream
- bytes already on the wire: typed openAIUpstreamStreamReadError, so the
handler appends an SSE error frame and marks the request an SLA failure
Client cancel/disconnect keeps the previous semantics — there the upstream read
error is indistinguishable from a truncation, so partial usage is still billed
rather than blamed on the account.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
normalizeOpenAIParallelToolCallsWithoutTools only looked at the top-level
"tools" array, but normalizeOpenAIResponsesLiteTools moves namespace tools
into an input item of type "additional_tools" and deletes the top-level key.
A Responses Lite request that carries tools therefore looks like it has none,
and the parallel_tool_calls:false that ensureOpenAIResponsesLiteParallelToolCalls
had just pinned gets deleted on the way out.
OpenAI then applies its default of true and rejects the request:
400 unsupported_value: "X-OpenAI-Internal-Codex-Responses-Lite requires
`parallel_tool_calls` to be false."
Note the field cannot simply be pinned to false unconditionally: without tools
OpenAI rejects it with "'parallel_tool_calls' is only allowed when 'tools' are
specified", so the two constraints have to be honoured together.
Reuse the same tool-detection standard the Lite path already uses by adding
openAIRequestBodyHasTools, the []byte counterpart of openAIResponsesLiteHasTools,
so both sides of the repo agree on what "has tools" means.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
The upstream names one offending index per response, but a replayed
conversation routinely carries dozens of items of the same type, each with a
status the upstream schema does not accept. Clearing a single index per round
trip needs one retry per item, so a conversation with more than
maxOpenAIResponsesRejectedFieldRetries such items exhausts the bounded budget
and the 400 reaches the client. Reported against tool_search_output items,
where the rejection surfaced as "Unknown parameter: 'input[60].status'".
Clear the status of every input item sharing the rejected item's type in the
same pass. Items of other types keep theirs: the rejection only proves that
the rejected item's type has no status field. When the rejected item carries
no type to match on, fall back to clearing the named index alone.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Add gateway.models_list_read_max_bytes with the existing 8 MiB behavior as its default, and apply it consistently to generic, Codex, and Antigravity model-list reads.
Read one sentinel byte for Codex manifests so oversized responses return an explicit bounded upstream error instead of malformed JSON.
- cc_pipeline: observe Chat Completions chunks/bodies as untyped payloads
(empty event type) so the upstream-echoed service_tier is trusted, matching
the upstream constraint that only terminal events and untyped bodies report
the actual processing tier.
- tests: add model field to terminal SSE frames (observation only triggers on
model-bearing frames) and assert response.created tier echo is ignored.
- openai_ws_forwarder_ingress: resolve billing tier from the local
upstreamResponseModelObserver (upstream echo first) instead of the raw
request payload, matching the HTTP->WS bridge and WS v2 forwarder.
- openai_ws_http_bridge_test: add fast-alias + upstream default case
proving the local observer's echoed tier wins.
- handler tests: keep only invalid service_tier -> 400 (short-circuits at
validation); valid/omitted semantics covered by the pure service-level
validation tests, avoiding real account selection in tests.
- Accept fast|priority (canonical priority), flex|auto|default|scale on
/v1/responses and /v1/chat/completions; reject unknown/empty/non-string
with HTTP 400; omitted and null stay compatible.
- Propagate service_tier through JSON/SSE, Responses<->Chat conversions,
fallback paths and HTTP->upstream WebSocket bridge.
- Billing prefers the upstream terminal tier; the outbound (policy-
transformed) tier is used only when upstream omits the field.
Explicit upstream default bills Standard even when Fast was requested.
- Pricing: Fast premium 2x Standard for gpt-5.6-sol/terra/luna and
gpt-5.4; 2.5x for gpt-5.5; channel FastMultiplier stays authoritative.
- Live verification (official Codex 0.149.0 + gateway, HTTP & WS):
upstream ChatGPT backend may return terminal default even when the
account catalog advertises priority; billing follows the actual tier.
DashScope/DeepSeek later tool_call deltas send empty id and
function.name. Clients that merge with !== undefined overwrite
the first delta's identity and dispatch unknown tool "". Drop
those empty fields on the raw Chat Completions SSE path.
DOMPurify <=3.3.1 (and the mermaid-transitive 3.3.3) carry ~18 disclosed
sanitizer-bypass/XSS advisories, including GHSA-cj63-jhhr-wcxv
(CVE-2026-65913): with USE_PROFILES enabled, ALLOWED_ATTR is rebuilt as a
plain array and looked up via ALLOWED_ATTR[lcName], so a polluted
Array.prototype property (e.g. onclick) is treated as an allow-listed
attribute and survives sanitization -- this app calls
DOMPurify.sanitize(svg, { USE_PROFILES: { svg: true, svgFilters: true } })
in src/utils/sanitize.ts, whose output is rendered via v-html in
ImageUpload.vue's SVG upload preview.
Bumped to 3.4.14 (latest, OSV-clean) and pinned via pnpm.overrides so the
mermaid-transitive copy dedupes to the same patched version instead of
staying pinned at 3.3.3. Lockfile-only regen via pnpm 9, no other package
changes.
A terminal event that arrives with an empty output was rebuilt from delta
accumulation. BufferedResponseAccumulator models only one reasoning item, one
message, and N function calls, and records no item id, status, or phase, so a
turn carrying several items collapsed into a single fabricated message: the
reasoning item disappeared, the real message id was replaced, and phase was
lost.
reconstructResponseOutputFromSSE already prefers the raw output_item.done
items over accumulation for buffered responses. The streaming path had no
equivalent because it never sees the whole body at once. Collect the raw item
of each output_item.done keyed by output_index and rebuild from those, falling
back to accumulation only when the stream reported no done item at all.
Items are stored as raw JSON, so vendor extensions and item types this gateway
does not model survive the rebuild verbatim.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Composite groups are a routing layer. The usage-side aggregation already
maps their traffic to the concrete account platform via
usageLogEffectivePlatformExpr, but the error aggregation read
ops_error_logs.platform verbatim. Pre-routing failures on composite groups
fall back to the group platform 'composite', which is never an enabled
monitor v2 config platform, so those error facts were filtered out of
every channel monitor v2 query and composite groups looked permanently
healthy.
Join groups/accounts in the error dedup CTE and resolve the effective
platform the same way as the usage side: account platform first, then a
real recorded platform, else 'unknown'. The window delete+rewrite
recompute rewrites historical facts with the corrected key.
Recognize GoUsageLimitError responses and parse both single and composite reset durations from their human-readable messages. Keep API-key accounts rate-limited until the parsed reset instead of falling back to the short default cooldown.
In Docker + cgroup v2 with no memory limit set, /sys/fs/cgroup/memory.current
returns a small container number while /sys/fs/cgroup/memory.max is "max".
readCgroupMemoryBytes then returned (used=<container>, total=0, ok=true).
collectSystemStats used that container "used" but, being unable to derive a
cgroup total, filled the total from the host via gopsutil. The dashboard then
computed container_used / host_total, e.g. ~60MB / 23GB ≈ 0.3% — wildly
understating real usage.
Fix: introduce resolveMemoryStats, which picks a single self-consistent
(used, total, percent) trio from ONE source. cgroup metrics are used only when
the cgroup exposes both a current usage AND a concrete limit (memory.max != max,
so total > 0); otherwise used/total/percent all fall back to the host reading.
The two sources are never mixed.
- memory.current valid + memory.max = "max" -> all host metrics
- memory.current = 512MiB + memory.max = 2GiB -> ~25% from cgroup
- no cgroup (bare metal) -> all host metrics
CPU metric behavior is unchanged (cgroup attempt then host fallback).
Adds ops_metrics_collector_memory_test.go covering all branches.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
The Responses client-tool restore raised an upstream function_call back to
custom_tool_call and tool_search_call without touching its item ID, so the
client received a custom_tool_call carrying the upstream's fc_ ID. That
combination is invalid: once the item lands in the client's history, any
later replay to an upstream that validates item IDs fails the whole request
with "Invalid 'input[N].id': 'fc_...'. Expected an ID that begins with
'ctc'", and every following message re-submits the same bad item.
Re-prefix the ID to match the type it is raised to, keeping the suffix so
the ID stays stable and unique per upstream item. IDs without a known
tool-call prefix are left alone rather than guessed at.
The lowering direction has to reverse this, because the WebSocket HTTP
bridge replays restored items back to the upstream and would otherwise lose
the upstream's own ID. normalizeLoweredFunctionItemID now maps a ctc_/tsc_
call ID back to fc_ instead of dropping it, which makes the round trip
lossless; output item IDs (ctco_/tso_) still have no function-protocol
counterpart and are still dropped. The helper was renamed from
dropInvalidLoweredFunctionItemID to match what it now does.
The stream restorer keeps the upstream ID for its own bookkeeping so later
upstream events still match the call, and emits the client-facing ID on
output_item.added/done and the custom_tool_call_input events.
Closes#6071
Account model mappings decide availability, but generated Codex catalogs previously inferred capabilities from local model-name tables. Compatible upstreams can expose unknown models such as OpenCode x-preview-f-free, so reasoning levels, image input, and context limits were missing or wrong.
Sync capability metadata from the account model endpoint, enrich incomplete lists from the Models.dev provider matching the account base URL, and persist only complete snapshots. Mixed groups consume the safe intersection across schedulable accounts. When IDs sync without complete metadata, return an explicit warning and preserve the previous snapshot.
Third-party Responses providers often omit /models. If that endpoint returns 404 or 405 and the account already has concrete model mappings, use those configured upstream IDs for capability enrichment. Authentication, rate-limit, server, and network failures remain hard errors, and metadata is never guessed across providers by model name alone.
Advertise a single none choice for non-reasoning models so current Codex can select them, then omit that catalog placeholder when forwarding to compatible Responses upstreams while preserving official OpenAI request semantics.
The admin user edit modal rejected concurrency < 1, so a user whose
concurrency is already 0 could not be saved at all — the guard runs
before the request, blocking notes, password, role and RPM edits on that
user too.
Everywhere else already treats 0 as unlimited: the gateway skips slot
limiting when maxConcurrency <= 0 (ConcurrencyService.AcquireUserSlot),
the batch limits endpoint binds concurrency with min=0, and the bulk edit
modal only rejects negative values.
Reject negative and non-integer values instead, mirror the RPM field with
min/step and a "0 = unlimited" placeholder and hint, and rename the error
key to match its new meaning. Account concurrency is unchanged.
The quick-add parser rejected every IPv6 proxy: the host group [^:]+
cannot match IPv6 literals (colons) and the pattern had no bracketed
form, so lines like socks5://[2001:db8::1]:1080 were reported invalid.
Add a bracketed-IPv6 host alternative and strip the brackets before
storing; the backend re-brackets via net.JoinHostPort when building the
proxy URL. Bare (unbracketed) IPv6 stays rejected because it is
ambiguous with host:port. Also add a regression test.
OpenAI groups can mix OAuth accounts with Responses-compatible API key accounts whose upstreams do not implement /models. The manifest handler previously selected one account and fetched its catalog before merging model_mapping aliases, so a permanent 404 returned 502 and hid models the administrator had already configured.\n\nTreat explicit model mappings from schedulable accounts as the authoritative Codex catalog. Build that catalog locally before account selection, preserve group filtering and final-body ETags, and retain the existing upstream discovery and failover path only when no mappings exist. Composite groups keep their existing account-first catalog behavior.
Anthropic fallbacks previously reused Antigravity defaults, leaking Gemini models into Claude-only groups. Composite aliases were also emitted with generic metadata and could be scheduled to accounts that did not own the exact mapping, while generated configs could default to a model absent from the downloaded catalog.
Keep provider defaults separated, derive alias capabilities from unique upstream targets with conservative fallback, require exact mapping ownership during scheduling, filter media targets, and choose generated config defaults from the fetched catalog.
Composite model catalogs expose exact account model_mapping aliases, but request routing only considered explicit routes or model-name detection. Unprefixed aliases were selectable but unroutable.
Resolve unique exact account ownership before the detector, keep explicit routes authoritative, and fail closed when an alias is claimed across platforms.
Codex treats the downloaded model catalog as a capability contract. Slug-only entries hide supported reasoning levels, can advertise invalid image inputs, and expose automatic or media-only models in the model picker. Generate complete route-aware metadata, preserve official manifests, filter non-agent models, and compute validators from the final catalog.
Codex model discovery expects a top-level models manifest, but Composite and other non-OpenAI groups either reached the OpenAI live-manifest handler or had no Codex-specific response at all. API key users also had no supported way to fetch that manifest and reference it from config.toml.
Generate minimal manifests from each group's effective model list while preserving the official OpenAI live path and the ordinary /models response. Add the Use Key flow for authenticated catalog download and model_catalog_json configuration without writing the API key into the downloaded file.
The model plaza route already supports public access, but both built-in /home headers omit its entry. Add the link to compact and default headers while keeping the existing feature and authentication settings authoritative, then cover the visibility matrix with focused component tests.
Constraint: Keep the change frontend-only and preserve router-owned access control
Rejected: Add the link to AppHeader only | /home renders its own headers and never mounts AppHeader
Confidence: high
Scope-risk: narrow
Reversibility: clean
Directive: Keep the model plaza entry gated by the existing opt-in flag and require-auth setting
Tested: HomeView focused Vitest, full frontend Vitest (223 files / 1554 tests), ESLint, vue-tsc, production build
Not-tested: Manual browser click-through against a running backend
Related: #5524
ConfigManager.refreshLoop reloads the Prompt Guard config every 5s and
Reload logged config_loaded on every successful load, so an unchanged
config produced up to ~17k identical lines per instance per day and
buried real configuration changes.
Log the event only when the reload carries news: the first snapshot, a
new config version (every admin save bumps it under the advisory lock in
UpdateConfig), a flip of the global risk control gate (a separate setting
that leaves the version untouched), or a recovery from a failed reload so
the degraded-to-healthy transition stays visible.
This mirrors logInvalidTokenEndpoints, which already warns once per
change rather than on every refresh.
CN-provider accounts explicitly configured with api_protocol=anthropic
(the frontend offers per-provider presets such as GLM Anthropic) fell
through the account-test router to the generic Claude tester, which:
- appended ?beta=true, which the native passthrough deliberately avoids
for third-party Anthropic endpoints, and
- defaulted a missing base_url to https://api.anthropic.com — sending
the provider's API key to Anthropic instead of the provider's own
Anthropic-compatible endpoint (guaranteed 401 plus cross-provider
credential exposure).
Route them to a dedicated probe that mirrors the adaptive Anthropic
check: resolve the endpoint via GetAnthropicProtocolBaseURL (same
resolution as real /v1/messages forwarding, including per-platform
defaults), use the shared API-key auth header, and mark 401/403 as the
account error state.
Also fail fast with an actionable hint when the anthropic-protocol
base_url still points at an OpenAI-compatible endpoint (paas path,
version segment, or chat/completions suffix): the naive {base}/v1/messages
join would 404 (e.g. /api/paas/v4/v1/messages) without any pointer to
the actual misconfiguration.
The quota cell used the noun label "5-hour window/weekly window" and the
balance cell the balance value itself as the click target for a manual
refresh. Both read as passive captions, so users could not discover the
manual probe and reported the feature as missing ("only OpenAI has a
refresh button").
- Split data display from the refresh action: bars/balance render as
static rows (snapshot already paints on mount), with a dedicated
action button below, aligned with the OpenAI "Query" and Grok
"Probe" buttons (same icon, blue action styling).
- Label the control with a verb (cnProviders.probe: 查询 / Query) and
give the balance cell a tooltip (it previously had none).
- Keep the tier/label layout classes and data-test hooks intact.
- Extend the quota cell spec and add a balance cell spec covering
snapshot rendering, explicit action labeling and failure passthrough.
sanitizeGroupMessagesDispatchFields forces AllowMessagesDispatch=false for
every non-openai platform including composite, and the gate checked only the
group platform — so composite requests resolved to grok/CN targets were
rejected 403 on /v1/messages and /v1/messages/count_tokens before reaching
the composite target whitelist, leaving the CN rollout unusable for
Claude-protocol clients.
- allowOpenAICompatibleMessagesDispatch: composite groups resolved to a
grok/CN target get the same exemption as the standalone platforms;
openai-resolved and unresolved targets keep requiring the switch
- resolveOpenAIMessagesDispatchMappedModel: skip the group-level dispatch
model mapping (openai-specific gpt defaults) for composite grok/CN
targets — model rewriting stays with account-level model_mapping
- Responses WebSocket composite whitelist stays openai+grok: CN accounts
cannot pass the WSv2 ingress transport filter and the WS HTTP bridge
has no Responses conversion for them, so admitting CN targets only
turns a clear policy rejection into a misleading no-available-account
- widen the responses/input_tokens and messages/count_tokens composite
gates to the shared openai-compatible text whitelist so CN targets get
the same local-estimate token counting as generation
- restore the Claude default-model fallback for standalone CN groups
(admin candidates and custom models list) while keeping composite
listings limited to CN account mapping keys
- refresh the scheduler bulk-rebuild comment and bucket capacity hints
for the 8-platform canonical set
The star history chart was broken due to GitHub stargazer API restrictions. Update the chart URLs in README.md, README_CN.md and README_JA.md so the chart renders again.
P2-2/P2-4/P2-5 from review:
- validateLinkedAccount/revalidateLinkedAccount now check
monitorAccountQuotaCapability after the platform match, blocking combos
that would permanently error at runtime: deepseek coding and
custom-domain kimi coding (no quota endpoint), zhipu payg (no balance
endpoint), anthropic/openai API-Key accounts (usage query requires
oauth; anthropic setup-token still allowed via local estimation).
gemini/grok/antigravity stay permissive. Quota mode surfaces
CHANNEL_MONITOR_ACCOUNT_NOT_SUPPORTABLE on edit; probe mode silently
unbinds (same as platform mismatch).
- applyMonitorUpdate re-runs validateCheckMode on the effective
provider+check_mode whenever either field is patched, so a
provider-only update can no longer persist antigravity+probe (the
check is conditioned on provider/check_mode presence so legacy illegal
rows can still be renamed/disabled).
- normalizeMonitorPrimaryModel only substitutes the "quota" placeholder
for pure quota mode; quota_probe with an empty model now fails with
MissingPrimaryModel instead of probing model="quota". The quota branch
also moves ahead of the grok default, so grok+quota now gets "quota"
rather than "grok-4.5" (review L-item).
Tests: capability matrix (15 cases incl. anti-over-blocking rows),
linked-account wiring, revalidate quota/probe split, provider-only
update bypass + legacy-row rename, quota_probe missing-model create,
normalize matrix.
P2-1/P2-3 from review:
- fetchCNQuota: credential-invalid now judged by StatusCode 401/403
(aligned with fetchCNBalance) instead of `!Success && !CredentialValid` —
CN quota service only sets CredentialValid=true on the success path, so
500/429/zhipu business errors were all misclassified as failed instead
of error.
- fetchCNBalance: snapshot carries new BalanceLow flag computed with the
scheduler's exact criterion (`!Available || allCNBalancesBelowThreshold`)
against Gateway.CNProviders.BalanceThreshold (ctor now takes cfg; wire
regenerated). quotaDegradedHint reports "balance low" instead of the old
`<=0` check, so an account already paused by the scheduler (balance 5 /
threshold 10) no longer shows green in the monitor.
- threshold helper falls back to viper default 0.5 for nil/<=0 config to
avoid a zero-threshold regression where balance=0 stops alerting.
Tests: CN quota status-code matrix (rewrites the test that cemented the
old behavior), balance-low matrix (below-threshold / unavailable /
multi-currency healthy), threshold-from-config; PayG stubs now set
Available explicitly (zero-value trap).
- ChannelsView platformOrder now includes the three CN provider
platforms so channel pricing can be configured for them; composite
group expansion/attachment stays limited to the five main platforms
(matches backend isConcreteRequestPlatform and composite-routes
target_platform validation)
- SyncPricingModels maps kimi->moonshot, zhipu->zhipu,
deepseek->deepseek; also fixes gemini mapping to "google" which
matched zero catalog entries (provider key is "gemini")
- Add CN platform colors to channel pricing model tag classes
- Replace ApplyCodexCanonicalIdentity with CodexCanonicalAuthIdentity /
ApplyCodexCanonicalAuthIdentity: the credential face (auth.openai.com
token exchange / refresh / PAT whoami) now sends the originator +
canonical User-Agent pair and no version header, matching codex-rs
default_headers(); the version gate (#3901) only exists on the
/backend-api/codex inference face. whoami keeps its original header
shape (originator + UA) with the canonical UA source.
- Token exchange and refresh send the full pair instead of a bare UA,
eliminating the half-identity (UA without originator) combination no
real client ever emits.
- Codex models manifest: the Version header now follows the client's
own client_version when it is valid and >= the upstream floor (same
source as the query param, restoring the pre-refactor consistency),
falling back to the canonical version otherwise; the query param
keeps its verbatim passthrough contract.
- Drop the now-unreferenced openAICodexProbeVersion constant and its
vacuous consistency assertions; probes resolve their version through
resolveCodexOutboundIdentity at runtime.
The struct field alone never reached the wire: the raw passthrough
pipeline is covered by enableMixedGeminiToolInvocations (#5711), but
TransformClaudeToGeminiWithOptions builds GeminiToolConfig from scratch
and never set the flag, so gemini-* models entering through the Claude
format gateway could still hit the upstream 400 from issue #5709.
- Set IncludeServerSideToolInvocations=true when the built tool
declarations mix functionDeclarations with googleSearch, matching the
raw-path injection semantics.
- Replace the marshal-roundtrip-only test with behavior tests that
drive TransformClaudeToGeminiWithOptions: mixed tools set the flag,
function-only and web-search-only requests leave it unset.
- user_repo.create(): keep the TxFromContext fast path, but restore
tolerance for dbent.ErrTxStarted in the self-owned-transaction branch.
ent's Client.Tx only inspects the driver type, so a repository built
from a tx-bound client (client-injected transactions, e.g. the
integration fixture testEntTx + tx.Client()) hits ErrTxStarted; reuse
that client instead of failing. Fixes the two red integration tests in
allowed_groups_contract_integration_test.go.
- createUserAndClaimInvitation: roll back via defer (matching the OAuth
registration precedent) so a panic inside the transaction cannot leak
the connection.
- settingRepoStub: guard call counters and state with a mutex; the new
concurrency regression test exercises it from multiple goroutines and
the unsynchronized counters were flagged by -race.
Reduce filtered admin usage statistics from four scans to one GROUPING SETS query so every breakdown shares the exact same filters. Add concurrent expression indexes for requested and upstream model filters on large usage_logs tables.
Token exchange, PAT whoami, models, probes, and pre-writes now follow
the same UA/version chain as Codex inference instead of hardcoded
codex-cli/0.91.0 or compile-time constants.
PR #1463 removed the Sora platform, but some references survived:
- README/README_CN/README_JA kept the 'Sora status (temporarily
unavailable)' sections and gateway.sora_* docs that the removal PR
never touched.
- deploy/config.example.yaml still documented ~130 lines of sora_*
gateway keys, the top-level sora: direct-client/storage block, and
token_refresh.sync_linked_sora_accounts - none of which map to any
field in the config structs anymore.
- The OIDC login PR (02a66a01c, branched off pre-removal main and
merged 4 days after #1463) re-added the dead
PublicSettings.SoraClientEnabled field, which no code ever sets.
- A release sync (748a84d87) re-introduced sora i18n keys that the
later i18n split (d9e514f98) faithfully carried into
locales/{zh,en}/admin/{overview,settings}.ts. No component references
any of these keys.
This drops all of the above. Pure deletions, no behavior change.
- Add IncludeServerSideToolInvocations field to GeminiToolConfig to prevent dropping client tool settings.
- Fix HTTP 400 error when mixing built-in tools (e.g. Google Search) with function calling on Gemini 3.6/3.7 models.
- Add serialization/deserialization unit test TestGeminiToolConfig_IncludeServerSideToolInvocations.
Fixes#5709
RegisterWithVerification checked CanUse() and then marked the code used in
two separate, non-transactional steps; the second step's failure was
swallowed ("invitation code mark failure does not affect registration").
Concurrent registrations with the same invitation code could all pass the
check and each create an account, turning a one-time invitation code into
an unlimited account factory (TOCTOU race).
Fix:
- AuthService: create user and claim the invitation code inside one DB
transaction (createUserAndClaimInvitation). The claim reuses
redeemRepo.Use's conditional UPDATE (WHERE status='unused'); losers are
rejected with INVITATION_CODE_INVALID and their transaction (including
the user insert) is rolled back. No-code registration path unchanged.
- userRepository.create: explicitly join an outer ent transaction via
TxFromContext instead of relying on Client.Tx returning ErrTxStarted
(ent's Tx never inspects the context, so the old reuse branch was dead
code and user inserts always committed in their own transaction,
leaving orphan users behind when the outer transaction rolled back).
Regression tests:
- unit: concurrent register with one invitation code must succeed exactly
once (8 goroutines -> 1 success, 7 x INVITATION_CODE_INVALID)
- integration: outer-tx rollback removes user and releases the claim;
commit persists both atomically
89d826be2 raised backend/go.mod to `go 1.26.6` and updated the three CI
workflows' version assertions, but left the Go builder image in all three
Dockerfiles pinned at 1.26.5. Since the official golang images set
GOTOOLCHAIN=local, the toolchain is not auto-downloaded and any image build
fails hard at `go mod download`.
CI does not catch this: the workflows build with actions/setup-go, not with
these Dockerfiles.
Also extend the Go-upgrade checklist in DEV_GUIDE.md, which listed only the
CI files -- that omission is why the Dockerfiles were missed.
Default codex_fingerprint_mode to off. v0.1.175 treated a missing key as
"session", so upgrading silently rewrote installation/session/thread/turn/
window identifiers for every existing OAuth account that had never configured
this field. The quota regressions in #5555, #5556 and #5582 line up with that
version boundary, with A/B reports that rolling back to v0.1.173 restores
quota. Convergence is now explicit opt-in (#5610).
Only accounts that never set the field change behaviour; explicit off /
device / session / full keep working exactly as configured. That required
flipping the persistence condition in all three account modals from
"!== 'session'" to "!== 'off'": the old rule deleted the key when it equalled
the default, which after the flip would have silently discarded an
administrator's explicit opt-in to session.
Also extend convergence to the passthrough path, which previously left client
identifiers untouched:
- resolve the ids once in forwardOpenAIPassthrough and rewrite
client_metadata on the raw bytes (gjson extract + sjson splice) because
passthrough is a hot path that must not fully unmarshal multi-MB bodies;
a shared core keeps the raw and map variants from drifting
- both request builders apply the staged ids at the same relative position
(after session isolation, before identity enforcement) so headers and body
share one id set and turn_id stays consistent
- stage the ids unconditionally, including nil: a failover from a converged
account to an off account must not leave the previous account's ids behind
OpenAI sunset the legacy unary /responses/compact endpoint (404, #5598,
#5624), so the account "compact probe" in the admin UI kept failing even for
healthy accounts, and the beta-feature negotiation header was only attached
to compaction turns.
Beta features (codex-rs session/mod.rs build_model_client_beta_features_header
+ client.rs build_responses_headers): the header is a session-level constant
attached to every /responses request, the WS handshake and /responses/compact.
Enumerating FEATURES shows no Experimental feature is enabled by default, so a
default install sends exactly "remote_compaction_v2". Mirror that:
- OAuth requests without a client-declared header get the default shape, so we
no longer produce a "header only on compaction turns" pattern real Codex
never emits (#5586 chains that strip the header)
- a client-declared header is preserved as-is: non-empty without v2 means the
user disabled the feature and the gateway must not rewrite that
- native v2 turns (compaction_trigger in body) always ensure v2 is present
- non-OAuth upstreams keep the compaction-turn-only behaviour
- the WS injection sits outside the client-header copy block so prewarm and
turn handshakes cannot land in different pool compatibility buckets
Compact probe now exercises native v2 (streaming /responses +
compaction_trigger) instead of the dead endpoint. Success requires an actual
compaction output item — scanning output_item.done/added, the terminal
response.output[] and the whole-JSON fallback — so a 2xx that silently drops
the trigger is reported as unsupported (the "got 0 items" class, #5478,
#5648). Probe identity is now UUID-shaped and applies the account's
convergence, matching real traffic on the same endpoint.
Codex captures x-codex-turn-state from /responses SSE, /responses/compact
JSON and the WS handshake (codex-api sse/responses.rs, endpoint/compact.rs),
then echoes it back on later requests of the same turn. The HTTP path dropped
it because the header is not in the generic response allowlist, while the WS
path already relayed it — an inconsistency that broke the protocol chain.
Relay it explicitly at every commit point instead of widening the global
allowlist (which would leak it into Anthropic/Gemini responses):
- streaming, non-streaming and SSE-to-JSON handlers relay it, clearing any
value left over by a previous failover attempt when upstream sends none
- under the first-output guard the header is only staged; provenance is
recorded when applyAttemptResponseHeaders actually writes it, because a
first-output timeout discards the staged headers and the client never
receives that blob
- record (api key + client session) -> minting account, TTL-bounded with an
opportunistic sweep, and strip echoes known to come from another account
before they go upstream. Stripping only: injection is the Claude bridge's
job. Same-account or unknown provenance passes through unchanged.
The responses->chat streaming converter re-emits the tool name as an empty
string on every function_call_arguments.delta because ChatFunctionCall.Name
has no omitempty. OpenAI-compatible clients accumulate the name from the
first tool-call delta, so the trailing "name":"" overwrites it and the
call fails dispatch (unknown tool ""). Empty names also break the next
round-trip upstream (missing field 'name').
Fixes#5631
The channel cache holds a groupID -> platform map with a 10 minute TTL, and
only channel Create/Update/Delete call invalidateCache(). Changing a group's
platform through the admin API therefore leaves the cache pointing at the old
platform for up to 10 minutes.
Channel pricing, model mapping and the model whitelist are all matched per
platform, so during that window the lookups silently miss: pricing falls back
to the global LiteLLM price list, renames stop applying and the whitelist
stops restricting. Nothing is logged.
Inject a narrow ChannelCacheInvalidator into the admin service (same shape as
the existing APIKeyAuthCacheInvalidator) and call it from UpdateGroup only when
the platform actually changed. The dependency is optional -- when it is nil the
cache simply rebuilds on TTL expiry, as before.
`validateNoConflictingModels` used `toModelEntry`, which only lowercases,
while `expandPricingToCache` keys the cache with
`normalizeChannelPricingModelName` (lowercase + TrimSpace + `.` -> `-`
for `claude-*`). Two pricing entries that the validator considers distinct
therefore collapse onto the same cache key, and the one written later
silently overwrites the other.
Add `toPricingModelEntry`, which reuses `normalizeChannelPricingModelName`,
and use it for pricing conflict detection. `toModelEntry` is left untouched
for model *mapping* validation, because `expandMappingToCache` keys the
mapping cache with plain `strings.ToLower` -- normalizing mappings the same
way would reject configurations that the cache keeps separate.
This completes the fix for #4754, which added the normalization to the
lookup side only.
Replace if-then-t.Fatal nil checks with testify require.NotNil so
staticcheck can prove the pointer is non-nil before field access.
Fixes SA5011 in payment_handler_test, parse_test, and
openai_images_incomplete_test.
Resolve conflict in backend/internal/handler/openai_gateway_handler_test.go.
main and this branch each appended a passthrough upstream stub plus a test at
the same two insertion points:
main openAIHTTPPassthroughSSERateLimitUpstream
TestOpenAIResponses_APIKeyPassthroughSSERateLimitUsesConfiguredPoolRetry
branch openAIHTTPPassthroughAuthFailoverUpstream
TestOpenAIResponses_APIKeyPassthroughPoolAuthFailureRetriesThenSwitchesToHealthyAccount
Both sides are kept verbatim; the only edit is giving each stub its own
calls() body instead of sharing the trailing one. No assertion was changed.
openai_gateway_passthrough.go and openai_oauth_passthrough_test.go merged
automatically.
The security audit group (Risk Control + Prompt Audit) was hidden from
the sidebar via `hideInSimpleMode`, but nothing else in the stack
restricts it in simple mode:
- `router/index.ts` simple-mode `restrictedPaths` does not cover
`/admin/risk-control` or `/admin/prompt-audit`
- the `/risk-control` and `/prompt-audit` admin route groups have no
`RunMode` gate, and neither does `internal/securityaudit/` nor
`service/content_moderation.go`
- `SettingsView.vue` renders the risk control toggle together with a
`<router-link to="/admin/risk-control">` shortcut, and the settings
page stays visible in simple mode
The feature is therefore fully usable in simple mode and already
reachable through the settings page — only the sidebar entry was
missing. Drop the flag so the menu matches actual behaviour.
The group remains gated by `risk_control_enabled` (opt-in, default
false) through `featureFlag: flagRiskControl`.
Three guards on the response_model billing basis, all scoped to the opt-in
channel mode so existing channels are unaffected.
1. Per-unit billing gate was stale. Audio (AudioUsage) and the search
surcharge (SearchCount) reached the billing paths after this branch was
cut; both are priced per unit rather than per token, so they must be
excluded like image/video/web-search already are. Audio pricing ignores
the model entirely, so the previous code "adopted" a basis switch that
changed nothing and emitted a misleading audit log for it.
2. Never zero out a billable request. A catalog entry whose token prices are
explicitly 0 still passes the identified-pricing gate (TokenPricingAbsent
only means both prices are missing), so an upstream could declare a free
model name and drop the bill to zero. Reject a zero (or negative)
recomputation whenever the baseline was billable; an already-zero baseline
is unaffected.
3. Never cross from channel pricing to the global table. Channel pricing
matches exact keys and prefix wildcards and does not strip date suffixes,
while the global table's identified lookup does. Upstreams routinely
declare dated model IDs (claude-opus-4-5-20251101), so allowing a
cross-source comparison would silently bypass an administrator's channel
markup on essentially every request. Admins who want a downgrade target
discounted can price it explicitly on the channel.
Also skip the recomputation entirely when the declared model equals the
baseline: it is provably the same cost and only burned a pricing resolve.
The identified-pricing helpers now return whether the model resolved to
channel pricing so the third guard costs no extra resolve.
DEV_GUIDE.md and the three READMEs still advertise Go 1.25.7 and
golangci-lint v2.7, but CI has since moved on:
- backend/go.mod declares go 1.26.5, and backend-ci.yml / release.yml /
security-scan.yml all resolve the toolchain via
`go-version-file: backend/go.mod` and then hard-assert
`go version | grep -q 'go1.26.5'`.
- backend-ci.yml pins golangci-lint to v2.9.
So a contributor following DEV_GUIDE.md installs a linter two minor
versions behind CI (different findings locally vs. in CI) and expects a
Go version that the workflow's own assertion step rejects.
Update all ten stale references, and note in the CI section which files
the Go version assertion lives in so future bumps don't miss one.
Docs only, no code or workflow changes.
PR #5423 relaxed the email suffix whitelist: once a whitelist is
configured, non-whitelisted registrable domains are each allowed to
register one account. That behavior activated unconditionally.
Add registration_email_domain_quota_enabled (default false) to gate it:
- Off (default): restore pre-#5423 strict whitelist semantics — with a
non-empty whitelist, non-whitelisted domains are rejected with
EMAIL_SUFFIX_NOT_ALLOWED; the register/verify views restore the
client-side whitelist pre-check and allowed-domain hint.
- On: keep #5423 behavior — one account per non-whitelisted registrable
domain (EMAIL_DOMAIN_REGISTRATION_LIMIT).
- Empty whitelist keeps allowing all domains in both states.
Gating lives in validateRegistrationEmailQuota and (as a race-safety
backstop) createUserWithRegistrationEmailGuard; the repository-level
domain lock + in-tx recheck is unchanged. The admin update field is
*bool (omitted = keep current) so stale full-payload saves cannot
silently flip the switch. Email binding and OAuth auto-signup keep
their strict policy, and pending-OAuth bind-login for existing
accounts is unaffected because the handler resolves existing emails
before the quota check.
Frontend adds the toggle to admin settings (zh/en copy; whitelist hint
restored to strict wording, quota wording moved to the new toggle) and
exposes the flag via public settings + SSR injection payload.
Tests: #5423 quota tests now enable the switch explicitly; new
default-off regression tests cover register/send-code/async/pending
OAuth/OIDC create-account plus both register views; API contract JSON
and the injection drift guard are updated.
Resolves three conflicts, all of the "both branches appended to the same
block" shape. Every one is resolved as a union of both sides; nothing from
either parent is dropped.
- handler/admin/setting_handler_update.go: keep ChannelMonitorHideThroughput
(V2) alongside GrokDefaultTextModel / GrokCrossClientModelMapEnabled /
GrokDefaultBaseURLMode (#5408). UpdateSettings writes every key on each
save, so dropping either side would reset those settings to zero values.
- service/domain_constants.go: keep SettingKeyChannelMonitorHideThroughput
and the three SettingKeyGrok* constants.
- repository/migrations_runner.go: keep the 195 checksum rule (V2) and the
218/219/220 rules (#5408).
Treat cacheTTL=0 as non-expiring known entries, always store negative
refresh markers, and update sticky getSchedulableAccount tests to expect
first-hit fail-open then block after background stats warm.
Keepalive is gated on downstream idle, not upstream tick cadence. The old
fixture wrote progress events every 250ms and could finish without a true
1s idle window on loaded CI runners, so ":\n\n" never appeared. Pause after
preamble long enough for the keepalive ticker before completing the stream.
Fix golangci unused/gofmt on the gentle-backfill path. Plot matrix and
line-chart X axes on the selected [requested_start, requested_end)
window (empty slots while backfill lags), and let plain mouse-wheel zoom
narrow the visible interval so pulse blocks grow wider.
H1/H2: bill search and voice with code defaults when group prices are
nil (explicit 0 remains free); bump API key auth snapshot to v19 and
refresh incomplete media/search/audio projections.
M1–M6: free-quota soft gate fails open on cache miss with background
refresh and 60s default TTL; correct password_auth config docs; default
cross-client model map to true (→ grok-4.5); audit /tts and /web_search;
exclude composite from migration 220 video-price clears; never let a
search surcharge mask token pricing failures.
A Responses stream that ends with response.completed but carries no output,
no usage and no error (only response.created + response.completed observed)
is a silent upstream refusal. It was previously treated as a successful
turn, recording 0 input / 0 output tokens without triggering failover, so
clients saw an empty reply and usage was never billed.
Detect the empty terminal event in both the standard and passthrough
streaming paths while nothing has been written to the client, and return an
UpstreamFailoverError so the next account is tried; when failover is
exhausted the client gets an explicit Responses-format error instead of an
empty success. Streams with any semantic output, usage or error are
untouched. Fixes#5009.
OpenAI upstream requires reasoning input item ids to begin with 'rs' and
rejects replayed item_* ids with 400 ("Expected an ID that begins with
'rs'."). shouldStripOpenAIResponsesInputItemID handled message (msg) and
tool call (fc) ids but not reasoning items, so long Codex sessions restored
through the OpenAI API key passthrough path failed with a converted 502.
Strip non-rs reasoning ids while preserving valid rs_* ids, mirroring the
existing message/tool-call handling. Fixes#5410.
Default channel_monitor_mode to v1 (opt-in V2) so upgrades keep active
probes; existing explicit v2 rows are left alone via ON CONFLICT DO NOTHING
plus migration checksum compatibility for already-applied 195.
V2 first-enable backfill no longer compresses ticks to 5s or uses 24h
chunks. Each tick does recent overlap plus at most one historical chunk
with depth-based ceilings (2h/4h/6h), adaptive grow/shrink, and failure
backoff. Error request_id dedup is bounded by a 90-minute lookback so
ops_error_logs is not scanned for full history.
Also align hide_throughput parse default with privacy-preserving public
runtime (missing key → true).
Check Close/CloseNow errors, drop unused helpers and dead constants,
lowercase ST1005 error strings, and stop discarding unwrap status as an
unused assignment so CI golangci-lint passes.
Expand Use Key Grok CLI config.toml to include production-oriented sections
from working gateway setups: full [endpoints], preferred_method=api_key,
image_description, auto_compact threshold, and Imagine image/video feature
overrides with guided comments.
Ship Use Key templates that match Grok Build / Codex best practice: env
vars + multi-model config.toml with api_backend=responses, env_key over
hardcoded secrets, and clearer shell/path guidance for Claude/Codex/OpenCode.
Also set free_quota_token_limit default to 500k (24h soft-gate), clean up
personal-dev-only comments, and keep billing test fixtures aligned.
Mint durable grok_audio/grok_realtime usage ids, avoid CLI headers on api.x.ai
voice, retry video pending store and fail-closed when snapshot is missing without
status duration, align pure-video ImageCount tests, and escalate unset search
price_per_1k to error-level logs.
Remove optional upload_url / fake connectivity-only success from admin video
tests. Stamp Grok CLI headers only on the CLI proxy so OAuth media against
api.x.ai can complete and preview video like the gateway path.
Show free accounts only the rolling 24h soft-gate bar; paid accounts show
7d/30d and prepaid money. Drop the always-visible manual probe chip from the
usage cell in favor of scheduled recovery and usage load.
Classify free-usage bodies during account tests without quarantining content
policy, and mark billing/spending-limit refresh failures as transient so
accounts stay probe-eligible.
Store CreatedAt on pending billing at video create and use wall-clock E2E
latency when status/content first observes official done+video.url, so usage
logs no longer record only the single poll hop.
Align gateway Grok media/voice paths and model lists, harden upstream failure
and quota handling, clear non-Grok video generation config migration, and polish
temp-unsched/status indicators with model whitelist updates.
Require oauth state/redirect consistency, fail closed on missing proxy,
and remove password login from create/reauth UI (admin-only password path stays off by default).
Add mode-first connectivity probes for text/image/video/search/tts/stt/realtime,
standalone voice and web-search paths, media upload options, and in-browser
image/audio/video preview (including ZDR-safe b64 images and edit validation).
- Change loadRuntimeSnapshot failure from fail-open to fail-closed
- Return block decision with 500 status when config cannot be loaded
- Add regression test for snapshot failure scenario
- Fixes#5388
- Add 'composite' to imagePricingPlatforms set in groupsImagePricing.ts
- Allow Composite groups to toggle allow_image_generation in admin UI
- Backend already supports allow_image_generation field for all platforms
- Add test case to verify Composite platform is included
Fixes#5369
When OpenAI Responses streaming has compact keepalive commit 200 headers
but no semantic SSE event has been written, upstream 4xx errors were
silently swallowed, causing clients to hang until timeout.
The fix distinguishes between 'headers committed' and 'meaningful SSE
output written'. Compact keepalive may commit headers for heartbeat
purposes without writing actual SSE events. In that case, the Responses
stream still needs its protocol-correct terminal response.failed event.
Preserves existing behavior when a complete upstream response or
meaningful SSE payload was already communicated, and preserves image
keepalive semantics.
Fixes#5289
#5261 added a captcha check to POST /auth/oauth/pending/create-account, but
the shared create-account form only gates its send-code button on the
Turnstile token — the submit button's disabled condition never included it.
Turnstile tokens are single-use, so handleSendCode resets the widget in its
finally block and clears the token. Submitting inside the window before the
widget re-solves omits turnstile_token from the payload, and the backend
answers ErrTurnstileVerificationFailed (turnstile_service.go:65). With an
interaction-required challenge the widget does not re-solve on its own, so
the failure persists until the user notices and verifies again — while the
button stays enabled and says nothing.
Gate the submit button on the token, mirroring the send-code button and
EmailVerifyView, which already gates its pending-OAuth submit the same way.
handleSubmit repeats the check because implicit form submission (Enter in a
text input) bypasses the button's disabled state.
The Tencent path is unaffected: handleSubmit already acquires a fresh proof
per submit, and turnstile_enabled is false in that configuration.
Verified with the component spec (11 passed) and vue-tsc --noEmit (clean).
The account+model transient breaker reset its failure streak whenever the
gap since the previous failure exceeded a one-minute window. That made the
breaker's sensitivity a function of request rate rather than upstream
health: a gateway called less often than once a minute never advanced past
streak 1, where the cooldown is zero, so a consistently broken account was
never blocked. Every request re-selected it, paid a full upstream attempt,
and only then failed over to a healthy account.
Observed on a low-traffic deployment: two accounts returning 500 and 503
stayed in rotation indefinitely, logging `failure_streak: 1, cooldown_ms: 0`
on every request and adding ~750ms to each one before a working account was
reached.
The streak is already cleared on success — recordSuccess deletes the entry,
and every OpenAI handler reports the schedule result — so the time-based
reset is not needed to recover a healthy account. Keep a TTL purely to bound
the map for account+model pairs that stopped being used, and raise it well
above the cooldowns so it no longer doubles as a streak reset.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Review follow-ups on the reset-credit caching flow:
- Recover account state BEFORE (and independently of) the reset-credit
display cache. A failed cache refresh could previously abort the run and
leave the account rate-limited — the very reason the credit was spent
(#3672 / #3740). The recovered account row is now returned even when the
cache refresh fails.
- Run the post-reset bookkeeping on a detached, time-boxed context and give
the panel reset call a larger timeout. A client abort no longer strands a
consumed (non-refundable) credit with an unrecovered account, and the
chained upstream calls can no longer exceed the client timeout and invite a
retry that spends a second credit.
- Persist the reset-credit snapshot through POST /accounts/:id/quota/refresh
instead of a side-effecting GET flag, so the write is covered by the audit
middleware. A rejected snapshot write now degrades to cache_persisted=false
instead of turning a successful upstream read into a 502 that left the card
without a credit count and the reset button permanently disabled.
- Reject snapshots whose positive count carries no expiration entries, and
drop expired credits (clamping the count) when rehydrating, so a stale
cache can no longer light up the reset button.
- Keep nil quota / rate-limit services nil in the handler's interface fields;
storing a nil *Service made the "not enabled" guards non-nil.
- Time-box the usage-refresh suppression and reuse handleAccountUpdated so the
patched row also enters the auto-refresh silent window.
AdminRefundDialog already implements a `requireForce` prop that renders the
force checkbox and blocks submit until it is checked, but AdminOrdersView
never consumed `require_force` from the refund response nor bound the prop,
so the checkbox could never render. Admins hitting a require-force case (for
example a user who spent their balance after requesting the refund) only saw
an error toast and had no way to complete the refund from the UI.
Keep the dialog open when the backend answers `require_force`, surface the
checkbox and the backend warning, and reset the flag whenever the dialog is
opened or closed.
Verified with `npm run typecheck` (vue-tsc --noEmit, clean).
Review fixes for the profit-control feature commit:
- Image intent no longer disables the profit gate. The shared /v1/responses
handler previously skipped the pricing context (and therefore the gate)
whenever the platform-wide image intent predicate matched, which includes
Codex's passive image_gen namespace declaration: any client could disable
admission control for anthropic/gemini/antigravity groups by declaring a
namespace tool in the request body. Both /v1/responses paths now always
install the token pricing context; image intent only drives capability
routing and image billing. Mixed token+image requests stay token-gated;
only dedicated media endpoints remain out of scope.
- Out-of-scope paths are now explicitly suppressed instead of implicitly
ungated: Grok media (billed by media multipliers; also prevents in-flight
video lookups from turning into spurious 404s), OpenAI-group count_tokens
(unbilled), and Live calls (duration billed) carry a suppress marker that
every install point honors, including the defensive scheduler-entry
install.
- The gate resolved during selection now travels back to handlers on the
AccountSelectionResult. The shared gateway installed the gate only on a
scheduler-local context, so handler-side post-slot terminal rechecks and
post-admission sticky binding were no-ops for anthropic/gemini/
antigravity/shared-grok requests (and for composite-routed member groups
on the OpenAI path). Handlers re-apply the carried gate via
ContextWithSelectionProfitGate before the terminal recheck and binding;
the WS acquired-selection branch gained the previously missing recheck.
- Sticky binding semantics restored for ungated traffic:
BindStickySessionAfterProfitAdmission falls back to the official eager
bind when no gate is installed (wait paths lost their only binding point
otherwise), reads the pre-existing binding at bind time only when gated
(removes the unconditional per-request Redis read the feature added to
the shared handlers), and the legacy engine's three selection-time
binding writes are skipped under a gate so a terminally vetoed account
can no longer become the new sticky target.
- Responses WS connections re-freeze pricingAt and re-resolve the gate at
every turn (BeforeTurn) and bill each turn with its own instant, closing
the connect-at-valley/bill-at-valley window; a turn that fails the
recheck closes the connection so the client reselects on reconnect.
- Terminal recheck no longer swaps a DB-fresh selected account for a
staler snapshot object (UpdatedAt guard), and observer counters are
documented as per-evaluation.
- Migrations renumbered 191/192 -> 192/193 after the passkey migration
landed upstream as 191.
New regressions: selection-carried gate propagation (control group proves
the pre-fix no-op), image intent not disabling the shared gate, eager
binding fallback without a gate (both services), gated
read-failure/sentinel-miss binding semantics, legacy-engine deferred
binding under a gate with eager behavior preserved ungated, turn-level
pricing refresh (config re-resolution, scheduled-group precedence,
suppress, mid-connection disable), and suppress-marker coverage for the
request pricing context.
Group pricing (rate multiplier, peak windows, per-user overrides) and
account cost (accounts.rate_multiplier) already live side by side, but
nothing stops the scheduler from handing a request to an account whose
cost multiplier exceeds what the group's pricing can profitably serve.
Add an opt-in per-group profit gate that filters scheduling candidates
by a margin rule, while ordering, scoring, stickiness and breakers keep
working unchanged among qualified accounts.
Admission rule: an account qualifies iff U <= D * (1 - min_margin -
safety_buffer) within a small relative epsilon, where U is
accounts.rate_multiplier (0 is legal; missing/negative/NaN/Inf are
conservatively rejected as invalid) and D is the requester's effective
downstream multiplier (user-group override ?? group default, times the
group peak factor) frozen at the request's pricing instant.
- groups gain profit_control_enabled / profit_min_margin /
profit_safety_buffer (migration 191); the durable auth-cache
invalidation trigger additionally watches the profit and pricing
columns (migration 192) so out-of-band group edits cannot leave
stale auth snapshots; GetByKeyForAuth explicitly projects the new
columns and the API-key auth snapshot version is bumped to force a
refresh of pre-existing snapshots
- request-level pricing instant: token entry points install pricingAt
into ctx; the profit threshold D and the RecordUsage peak factor
read the same instant, so one request never changes price mid-flight
across waits/retries/failover (media and unwired paths keep the
existing record-time semantics)
- the gate covers token requests on openai, anthropic, gemini, grok
and antigravity groups: OpenAI-family handlers via
WithOpenAIRequestPricingContext (responses incl. WS bridge, chat
completions, messages, embeddings, alpha search), the shared gateway
via WithGatewayTokenRequestPricing (messages, chat completions,
responses, gemini model actions); composite groups cannot enable it
directly; image/video/models/usage/count_tokens stay ungated and an
explicit image-generation intent suppresses the gate end to end
- post-slot recheck: after a slot is acquired the account is re-read
via SchedulerSnapshotService.GetAccount (scheduler cache, then DB;
only when both fail the check fails open with WARN + metric); a
vetoed account releases its slot and joins the request's exclusion
set for reselection; sticky bindings are written only after the
final check passes, and an over-threshold sticky account is skipped,
not deleted, so it comes back once its rate recovers
- sticky-session cache contract: GatewayCache.GetSessionAccountID now
returns ErrStickySessionNotFound on a miss (mapped from redis.Nil in
the repository implementation, mirroring ErrRefreshTokenNotFound) so
the profit sticky path can distinguish "no binding yet" from a real
read failure without importing the cache driver in service code
- cross-group re-entry (composite parent -> member group) resolves the
gate against the member group and clears a stale parent gate instead
of letting a foreign threshold veto accounts
- per-platform/group activity counters (installs, threshold vetoes,
invalid-rate vetoes, refresh failures) for observability
- admin UI: profit-control section on the five platforms' group forms
with percent input, validation and platform-switch reset; group
create/update/duplicate normalize and validate the config at a
single choke point
- cmd/profit-preview: offline what-if tool that replays the production
admission semantics over an exported config/account/override/model
dump, reports per-model admitted-account counts under the default
and the worst-case (lowest user override) D, and surfaces probe-sync
staleness as warnings without affecting admission
Tests: service unit coverage for gate resolution/veto/threshold
epsilon/pricing instant/suppress marker/scheduler filtering and
post-slot recheck (incl. -race on the profit surface), unit-tagged
handler slot-recheck and capability-mapping regressions, sqlmock and
real-PostgreSQL integration regressions for the GetByKeyForAuth
projection and the migration-192 trigger watch list, API contract
update, and frontend specs for the five-platform form helpers.
Successful upstream billing probes already persist the upstream-declared
rate as a display-only snapshot. Add a per-account opt-in that writes
that declared rate back to the account's rate_multiplier, so the account
cost basis follows upstream repricing automatically instead of drifting
until an operator notices.
- new per-account flag upstream_billing_rate_sync_enabled stored next to
the probe flag in account extra: enabling sync force-enables the
probe, disabling the probe cascades sync off, and eligibility follows
IsUpstreamBillingProbeIdentity (tightened from any non-empty platform
to an explicit whitelist of the five supported API-key platforms so
future platforms do not silently inherit probe/sync semantics)
- only a successful probe whose declared rate survives validation
(finite, within bounds, not rounded to zero at the rate_multiplier
decimal(10,4) scale) writes back; failed/unsupported/invalid probes
leave rate_multiplier unchanged
- the writeback rides the existing snapshot CAS transaction:
UpdateUpstreamBillingProbeSnapshot takes an optional rateMultiplier
and applies it atomically with the snapshot under the same
identity/snapshot compare-and-swap, so a probe result observed on a
stale account cannot clobber a concurrent admin edit
- admin edit goes through UpdateWithAccountBillingSettings, which
applies the form without overwriting a rate that a probe synchronized
after the edit form was loaded (nil rateMultiplier = not edited);
once sync is enabled the edit form shows the rate as managed
- bulk update rejects a manual rate_multiplier change when any target
account has rate sync enabled (whole batch fails with a dedicated
error so partial writes cannot bypass the sync ownership)
- frontend: sync toggle with hints in the edit modal (probe/sync
enable/disable coupling enforced in the form), synced-rate tooltip on
the rate cell, bulk edit modal warns and blocks rate edits that hit
sync-enabled accounts; en/zh copy updated
- tests: service unit tests for sync gating/validation/cascade, sqlmock
repo tests for the extended CAS, real-PostgreSQL integration tests
(rate written only for successful+enabled accounts, manual rate
protected after sync disabled, admin edit preserved across concurrent
probe sync), handler/API contract updates, frontend specs for modal
coupling, bulk rejection and rate cell
coder/websocket arms a context.AfterFunc that hard-closes the connection
when a write context is canceled, and AfterFunc stop does not wait for a
callback that already started. An external cancellation (e.g. ingress
lease loss) landing inside the disarm window of an already-successful
downstream write could therefore kill the TCP connection before the
retry close frame (1013) was written, leaving the client with a bare
EOF. Mirror the read side: bound downstream writes with the write
timeout only, and rely on the explicit Close/CloseNow performed by every
relay exit path for teardown.
Fixes the flaky TestPassthroughLifecycle_LeaseLossSendsRetryClose.
/v1/sub2api/billing is a key-scoped sub2api convention: any API-key
account whose base_url points at a sub2api-compatible upstream answers
it regardless of the account platform. Widen probe eligibility from
platform=openai to every API-key account (OAuth/Bedrock stay excluded:
no static key to present).
- central predicate exported as IsUpstreamBillingProbeIdentity; runner,
manual probe, SetAccountEnabled, admin create/update/bulk validation
and CRS reconcile all follow it
- probe target resolution reads credentials.api_key/base_url directly.
OpenAI keeps its official-default base URL and openai transport
profile. Other platforms whose base_url is empty or points at an
official provider API domain (anthropic.com, googleapis.com, x.ai,
grok.com, openai.com - matched on the normalized hostname, port and
trailing dot stripped, as the exact host or any subdomain) persist
"unsupported" without sending a request: the create form fills empty
base_url with official defaults and offers official regional presets
(e.g. us-east-1.api.x.ai), and official APIs cannot answer
/v1/sub2api/billing, so probing would only send the account key to a
nonexistent official path
- due-scan SQL, BulkUpdate probe WHERE, UpdateCredentials stale-snapshot
CASE and proxy-change invalidation drop their platform filters
(type='apikey' retained)
- frontend: rate cell, edit/create/bulk modals gate on type==='apikey';
the antigravity upstream create flow (its own helper and form section)
shows the auto-probe toggle and passes upstream_billing_probe_enabled;
bulk WS-mode section stays OpenAI-only; settings copy de-scoped
- tests: multiplatform service unit tests (relay success, official/empty
base_url unsupported without request, normalized official-host matrix,
OpenAI defaults preserved), real-PostgreSQL due-scan coverage for
openai/anthropic/grok plus oauth/disabled exclusion, updated
sqlmock/CRS/frontend specs; probe stays opt-in per account
Fixes#5148: with aggregator upstreams (e.g. newapi) that end SSE
streams without a proper terminal event, every such request was
silently missing from usage logs and billing.
Root cause (tracked via the nested audit issue): the low-level
Anthropic SSE readers already return the partially collected usage
together with the stream error (missing terminal event, read error,
interval timeout), but Forward converted every such result to
(nil, err) and the handler returned before submitting RecordUsage.
Changes:
- Add partialStreamUsageResult: on stream errors, wrap observed usage
into a ForwardResult and return it alongside the error, for both the
regular Anthropic path and the API-key passthrough path. Invariants:
UpstreamFailoverError always keeps result=nil (failover retries are
billed as the successful attempt, never twice), and zero observed
usage returns no partial result (no phantom zero-usage records).
- Messages handler: hoist the usage submission block into a closure
shared by the success path and the new partial-result error path.
- Usage record worker pool: distinguish pool-stopped drops
(dropped_stopped) from operator-configured drop/sample overflow
drops; billing tasks now fall back to inline synchronous execution
only during the shutdown window, while explicit drop/sample overflow
semantics are preserved. Image usage keeps its mandatory fallback
for both drop kinds via the new mode.Dropped() helper.
Tests: Forward-level regressions for missing-terminal / read-error /
no-usage / failover-invariant on both paths, plus handler-level
stopped-pool sync fallback and drop-policy preservation tests.
Real claude-cli/2.1.220 auto-mode classifier requests carry two system
entries: the security-monitor prompt plus an appended session-context
block. The previous len(systemEntries) != 1 guard rejected them before
any content check ran, so claude_code_only groups kept refusing the
classifier (#5152, follow-up to #5041/#5048).
Scan every entry for the monitor prompt instead of requiring exactly
one. Discrimination is unchanged: the matching entry still needs the
10k-char minimum, the fixed prefix, and all eight markers.
OpenAI Responses may return code=rate_limit_exceeded in a response.failed SSE event while the HTTP status remains 200. Classify these failures as 429 so configured pool-mode retries and account failover are applied. No account-level rate-limit state is written on this path: the 200-stream response headers carry normal quota snapshots, and retry semantics stay owned by the failover engine.
Ignore compact keepalive bytes when determining whether semantic output has started, preserving safe retries before real output.
Implements #2646: the risk-control content audit can now send OpenAI
Moderations requests through a proxy from IP Management - Proxy Servers.
Backend:
- ContentModerationConfig gains proxy_id (nil = direct, unchanged default)
- update semantics: null keeps, 0 clears, >0 selects (validated to exist)
- moderation calls build the client via the shared httpclient pool; proxy
resolution failure surfaces as a moderation error and never silently
falls back to direct connection
- proxy_id -> URL resolution cached 60s (single-entry, invalidated on
config save) so the pre-block hot path does not hit the DB per request
- test-key endpoint accepts proxy_id too (null = saved config's proxy,
0 = force direct), so input-key/saved-key tests exercise the same path
- proxy usage/inactivity logged (content_moderation.proxy_enabled /
proxy_not_active) without leaking credentials
Frontend:
- ProxySelector in the risk-control basic settings tab, proxy list loaded
non-blockingly; save and test payloads carry proxy_id; zh/en i18n
- UseTLS now tries implicit TLS first (port 465 semantics) and, when the
server answers in plaintext (tls.RecordHeaderError, e.g. port 587
submission), automatically retries with mandatory STARTTLS; encryption
is never silently downgraded (fixes#1470, supersedes #1488)
- TestSMTPConnectionWithConfig now shares connectSMTP with the send path,
adding the opportunistic STARTTLS upgrade the send path gained in
b402c367d; this removes the 'test connection fails but test email
sends' mismatch reported in #1488
- test-connection now also honors dial/IO timeouts and ignores
non-standard QUIT responses, matching the send path
Reason:
- Responses retries can carry account-bound encrypted compaction items that OpenAI rejects with invalid_encrypted_content.
Changes:
- Drop encrypted compaction and compaction_summary items only during the existing recovery retry.
- Preserve unencrypted compaction items and cover HTTP and WebSocket recovery paths.
glm-5.2 has no entry in the fallback table, so getFallbackPricing falls
through to `strings.Contains(modelLower, "glm-5")` and prices it at
GLM-5 rates ($1.00 in / $3.20 out per MTok) instead of the official
z.ai rates ($1.40 / $4.40) — roughly 27% under.
LiteLLM carries no bare `glm-5.2` key either (only provider-prefixed
`cloudflare/@cf/zai-org/glm-5.2` and `fireworks_ai/.../glm-5p2`, which
the lookup candidates never match), so the request always lands on the
fallback path and the discrepancy shows up directly in usage logs.
Add the glm-5.2 entry (same price as glm-5.1 per docs.z.ai) and match it
before the bare `glm-5` branch, with a note that dotted variants must
precede it. The existing regression test asserting the old glm-5 price
is updated accordingly.
Source: https://docs.z.ai/guides/overview/pricing
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
The count_tokens path now strips max_tokens; this sibling test also
exercises ForwardCountTokens but still asserted max_tokens preservation.
Flip its assertion to expect the field is filtered, matching
CountTokensFiltersGenerationFields.
- Move pool mode check before the status switch so 401/402/403/5xx
all skip tempUnschedule consistently
- Skip rateLimitGrok in updateGrokUsageSnapshot for pool mode
- Expand test coverage to all affected status codes
The proxy stream circuit introduced in v0.1.164 (#4749) removes every
account behind a quarantined proxy from scheduling. When all schedulable
accounts share one proxy (a common deployment), two mid-stream
disconnects within a minute zeroed out capacity for 10 minutes and every
request failed with 502. One HTTP/2 connection loss also killed all
multiplexed streams at once, tripping the threshold from a single event.
- Quarantine now degrades to a preference: when the only reason no
account is available is proxy quarantine, selection retries once with
the quarantine bypassed, so capacity can never reach zero.
- Disconnects within 3s per proxy collapse into one failure event.
- Add gateway.openai_proxy_stream_circuit.disabled escape hatch.
- A completed stream still clears the quarantine immediately; TTL,
thresholds and recording guards are unchanged.
- Add compact_home_enabled setting to provide a minimal landing page
- Preserves custom home_content priority over compact mode
- Renders only site identity, navigation, and login/dashboard link
- Avoids marketing copy that triggers anti-fraud systems
- Includes focused backend/frontend tests and i18n support
Fixes#5065
Real claude-cli/2.1.220 samples insert <category>...</category> between
</block> and <reason>, breaking the previous literal marker.
Also fix an unchecked type assertion in the test suite.
`cli-chat-proxy.grok.com` turns away requests that do not name a
supported client version — the failure mode #3952 was opened for. The
pin has sat at 0.2.93 since 2026-07-17, while https://x.ai/cli/stable
and `npm view @xai-official/grok version` both report 0.2.114.
The version was written out in three places, which is how they came to
drift:
- `repository/http_upstream.go` (OAuth traffic through the CLI proxy)
- `pkg/xai/billing.go` (billing and quota probes)
- `service/openai_gateway_grok.go` (gateway request headers)
`xai.CLIClientVersion` is now the single source, and the other two build
their own client identity from it, so the next bump is one line.
`internal/pkg/xai` is a leaf package both layers already import, so this
adds no cycle. `CLIUserAgent` derives from the same constant.
Checked against the real 0.2.114 binary rather than the version feed
alone: the `x-grok-client-version` header name, the `xai-grok-cli`
token-auth value, and the version string all match what the CLI sends.
The pinned value is also the floor an operator override must clear, so
some fixtures carried a meaning relative to it rather than a fixed
number. Those were moved, not search-and-replaced:
- the "valid override" case now uses 0.2.115-alpha.1; at 0.2.95-alpha.1
it would fall below the new floor, be dropped, and stop testing
acceptance at all
- the prerelease-at-the-minimum case tracks the pin to 0.2.114-beta.1,
or it becomes a copy of the below-minimum case
- three of the five malformed-semver entries sat below the new floor and
would have been turned away for being old, not for being malformed
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Expand visible Composite groups into each configured concrete model platform while preserving ordinary group isolation and empty-state behavior.\n\nFixes #4985
The PASSKEY_DISABLED silence guard compared the string error code
against error.code, but the api client puts the numeric envelope code
there and the string code in error.reason, so the guard never matched
and every /profile visit on deployments without WebAuthn configured
showed a spurious "failed to load passkeys" toast.
Read error.reason instead, and skip the credentials request entirely
when the feature is disabled so the card no longer issues a request
that is guaranteed to fail with 403.
UserRepository.Update and APIKeyRepository.Update rewrote the whole row on
every call, regardless of which fields the caller meant to change. Several
columns on those tables are maintained by dedicated atomic paths (balance
deduction, quota and rate-limit counters, limit adjustments, activity
timestamps), so a caller holding a slightly older snapshot could silently
roll them back - a lost update.
Both methods now take an explicit column mask and persist only the columns
the caller declares; everything else keeps its current database value.
- All user and API-key call sites declare exactly what they mutate, which
turns admin edits and profile saves into genuine partial updates.
- Email uniqueness locking/lookup and allowed_groups sync only run when
those fields are part of the update.
- UserUpdateFields deliberately has no balance/total_recharged members, so
Update cannot touch them. New AdjustBalance/SetBalance apply the change in
a single statement and return before/after values; admin balance
adjustment uses them instead of read-modify-write.
- promo_codes.used_count is no longer written by Update; it is only ever
incremented by the redemption path.
- The billing hot path that marks an API key quota-exhausted writes only
status.
- Dropped a no-op row write in RevokeAllUserTokens: users has no
token_version column, so it persisted nothing while still overwriting
concurrently-updated columns.
Adds integration coverage that a stale snapshot cannot revert concurrent
atomic writes, and unit coverage pinning the column set each entry point
declares.
A hijacked session must not be able to silently add a passkey as a
persistent backdoor or remove the victim's credentials. Registration
(begin) and deletion now verify the account password server-side,
reusing the existing PASSWORD_REQUIRED / PASSWORD_INCORRECT errors.
The password is used instead of TOTP step-up so the guard also protects
deployments that never configured a TOTP encryption key. The password
key in both request bodies is covered by the audit middleware's
key-substring redaction, so no credential material reaches audit_logs.
Frontend: the add-passkey form gains a current-password field, and the
delete confirmation is now a dialog with a password input (replacing
window.confirm), mirroring the TOTP disable dialog. Backend error
messages (e.g. wrong password) are surfaced instead of the generic
failure toast. Rename remains password-free as it is cosmetic.
- parseSettings now reports passkey_enabled=false whenever the WebAuthn
deployment config is absent: a stale "true" row left behind after the
config is removed previously made the admin update gate reject every
settings save while the UI toggle was disabled, leaving no recovery
path from the admin panel. Added a regression test.
- update the admin settings API contract goldens with the new
passkey_enabled/passkey_configured/passkey_rp_id/passkey_rp_origins
fields.
- errcheck: check rows.Close in passkey repository (repo convention).
- staticcheck QF1001: apply De Morgan's law in WebAuthn origin scheme
validation.
Administrators often need exact model identifiers while editing account whitelists. Add a dedicated copy action without changing model selection or upstream sync behavior.
Constraint: Keep the contribution frontend-only and avoid model routing or persistence changes
Confidence: high
Scope-risk: narrow
Reversibility: clean
Directive: Keep copy and selection as separate actions
Tested: focused Vitest, account component regression tests, frontend typecheck, lint, and production build
Not-tested: Authenticated browser screenshot
Related: Wei-Shaw/sub2api#2151
Flattening Codex `type:"namespace"` tool declarations into `namespace__name`
renames the tool but cannot rewrite the contract the client already handed the
model: Codex tells it to call `to=functions.collaboration.spawn_agent`. Under
`tool_mode: code_mode_only` (gpt-5.6-*) collaboration tools are the model's only
direct channel, so the rename left it unable to address them — it fell back to
`functions.exec` and looped on meaningless shell commands until the session
collapsed, with no upstream error anywhere. Fixes#4978.
Preserving is what the upstream actually wants. OAuth egress is unconditionally
chatgpt.com/backend-api/codex/responses (base_url is read only for API-key
accounts), i.e. the party that defines the extension; codex-rs builds one
request for both WS and HTTP and falls back WS->HTTP mid-session, so these
declarations already reach that host over HTTP — as they do today via the
responses-lite carrier and the WS/HTTP bridge, neither of which ever flattened.
Namespace names cannot be allowlisted: `features.multi_agent_v2.tool_namespace`
is user-configurable (openai/codex#31864 recommends renaming it to `agents`) and
MCP/connector namespaces are generated at runtime (mcp__codex_apps__gmail).
So the default is inverted rather than extended with more names.
- Preserve namespace declarations by default for OpenAI OAuth; the new
per-account `extra.openai_responses_flatten_namespaces` restores the previous
behavior for deployments routing OAuth traffic to a relay that rejects them.
- Keep `input[].namespace` on tool-call items for OAuth non-compact requests.
The upstream requires the round-trip ("Missing namespace for function_call
'...'. Round-trip the model's function_call item with its namespace field
included."), so preserving declarations while stripping the calls would 400 on
the second turn. The item-type allowlist mirrors the existing reactive strip.
- Keep compact on its current behavior: that endpoint rejects the field outright
("Unknown parameter: 'input[894].namespace'", #4761) and there is no evidence
either way about namespace declarations there, so this change does not widen
its surface. API-key accounts are untouched — their upstream is a standard
Responses API and the reactive retry only clears one index per round trip.
- Clear the flatten mapping at the start of each Forward attempt so an account
reached through failover cannot restore responses with the previous account's
mapping.
Admin UI gets an OAuth-only toggle (create/edit/bulk) plus zh/en copy noting the
compact exception.
An OpenAI account with auto-passthrough enabled (extra.openai_passthrough=true,
"replace auth only, allow all models") was still filtered out during account
selection when it had a non-empty credentials.model_mapping that did not list the
requested model. isOpenAICompatibleAccountEligibleForRequest calls
Account.IsModelSupported directly, and IsModelSupported only bypassed the mapping
whitelist for passthrough accounts when the mapping was empty. A leftover mapping
(common after switching an account from whitelist mode to passthrough) therefore
excluded the account -> zero candidates -> ErrNoAvailableAccounts, and the client
saw 404 "Model ... is not supported by any configured account in this group" even
though a direct account test with the same model succeeded (that path already
honored passthrough via isModelSupportedByAccount).
Fix: short-circuit passthrough at the top of Account.IsModelSupported, before the
model_mapping check, so it is consistent with isModelSupportedByAccount. Add a
regression test covering passthrough + non-empty leftover mapping.
When an upstream API gateway (e.g. new-api) relays real Claude Code
requests, the User-Agent becomes Go-http-client while the body retains
the full Claude Code fingerprint (billing attribution block +
metadata.user_id + cache_control breakpoints).
Previously, the OAuth mimicry path relied solely on UA matching to
detect Claude Code clients. Without a matching UA, the gateway would
rewrite the system prompt — replacing the client's carefully structured
system blocks and cache_control breakpoints with its own injection.
This breaks Anthropic's prefix-based prompt cache: since the cache key
evaluates tools → system → messages in order, a changed system
invalidates all downstream message caching.
Symptoms observed:
- cache_read permanently locked at ~25K (only system prompt cached)
- cache_creation growing monotonically every turn (full messages rewrite)
- Single-request costs $17-27 instead of normal $1-2
Fix: when UA does not match but the body contains a valid billing
attribution block (x-anthropic-billing-header with cc_entrypoint=),
treat the request as proxied Claude Code traffic and skip mimicry.
This preserves the client's original system structure and cache_control
breakpoints, allowing Anthropic's prompt cache to function correctly.
Manual Grok connection tests previously surfaced upstream HTTP 402 errors without changing account availability. Persist the same 30-minute payment-required cooldown used by the live forwarding path so refreshed account lists no longer present the account as schedulable.
Constraint: Keep manual-test HTTP 402 handling aligned with existing Grok forwarding semantics.
Rejected: Mark the account permanently error | payment state can recover and the forwarding path intentionally uses a bounded cooldown.
Confidence: high
Scope-risk: narrow
Directive: Keep the manual-test cooldown reason and duration aligned with handleGrokAccountUpstreamError.
Tested: go test -tags=unit ./internal/service -count=1; go vet -tags=unit ./internal/service; production package compile check
Not-tested: Live xAI account with an exhausted subscription
Related: #4794
When a /responses forwarding loop has attempted an OpenAI passthrough
account, sanitize every subsequent non-passthrough attempt by deriving its
body from the immutable canonical request and dropping provider-specific
encrypted reasoning input items in full.
This prevents Bedrock-compatible accounts from rejecting Kiro reasoning
IDs and encrypted_content, including on same-account retries and later
non-passthrough failovers. Preserve JSON numbers exactly while sanitizing.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Go persists upstream_billing_probe.next_probe_at via RFC3339Nano, but
jsonpath datetime() parses at most 6 fractional digits, so every stored
timestamp failed to parse and was treated as malformed: fail-open due,
ordered into the invalid bucket by id ASC. With more enabled accounts
than the per-cycle limit, the same lowest IDs monopolized every cycle
and higher IDs were never probed again. Trim the fraction to
microseconds before datetime(), mirroring ListDueOllamaCloudUsageAccounts,
and pin the behavior with integration regressions: nanosecond parsing,
due-time ordering beyond the limit, preserved fail-open for truly
invalid dates.
All four pass on a quiet box and fail on a loaded one, each for its own
reason. None of them is testing the clock, so none of them should be
failing on it.
- ollama_cloud_usage_test.go: the second caller was released with
`close(release)` right after its goroutine was started, not after it
had reached the singleflight group. When the first refresh won that
race, the second became a new singleflight execution, re-read the
account, saw the LastAttemptAt the first one had just written, and
came back with the 30-second manual-refresh 429 at line 685. It now
counts account loads and waits for the second caller's own load,
which happens right before it joins the group. Adds a counting
GetByID to the test repo.
- gateway_hotpath_optimization_test.go: a 20ms sleep was meant to let
all 12 callers reach the cache before the loader was released. A
caller that arrived after the load had finished got a hit, not a
miss, so the miss count came out 11 of 12. It now waits on the miss
counter itself, which is the value the test asserts on.
- token_refresh_pool_health_test.go: the floor was `configuredSpacing`
minus 10ms, i.e. 40ms out of 50ms. Each start timestamp is taken
after the rate gate releases the goroutine, so scheduler delay can
compress one observed gap with the gate behaving correctly — seen at
37ms and again at 13ms. The floor is now a tenth of the configured
spacing. Measured with providerQPS=20, 8 attempts, concurrency 2:
gate at 50ms gives a minimum gap of 49.97ms, gate at 0 gives 22µs.
So an unpaced gate sits three orders of magnitude under the 5ms floor
and is still caught, while jitter has room to move. A comment warns
against replacing this with an assertion on the total span of the
starts: the span is set by how long each attempt takes under the
concurrency limit, not by the gate — 471ms paced against 241ms
unpaced — so a span check passes with the gate disabled.
- prompt_guard_test.go: the bound only has to show the failover shared
the first endpoint's 70ms deadline and did not take the second
endpoint's own 500ms one. An unshared deadline lands near 535ms, so
350ms still fails loudly (seen: 224ms against a 180ms bound).
Tests only; no product code is touched. Each fix was checked in both
directions: it passes with the behaviour intact, and it still fails when
the behaviour is broken on purpose (for the QPS one, by swapping the
shared rate gate for a zero-interval one).
PUT /api/v1/admin/settings is a whole-document write. The admin UI always sends
the complete document, so saving from the settings page is unaffected both
before and after this change. The bug is only reachable when an API client calls
the endpoint directly and sends just the fields it wants to change, which is the
natural assumption for a PUT on a settings resource.
Value-typed fields of UpdateSettingsRequest bind to their zero value when the
payload omits them, and buildSystemSettingsUpdates writes every key
unconditionally, so such a caller has no way to say "leave this one alone".
Omitting a field and explicitly clearing it are indistinguishable on the wire.
A caller that sends only the field it wants to change, e.g.
{"risk_control_enabled": true}
sets that flag and clears every other unguarded field in the same request.
Measured against a fully configured store, one such call empties site_name,
site_subtitle, api_base_url, contact_info and doc_url, and turns
registration_enabled, email_verify_enabled, invitation_code_enabled and
turnstile_enabled off. turnstile_enabled alone gates the captcha on login,
register, forgot-password and both verify-code endpoints, and
email_verify_enabled is a precondition of IsPasswordResetEnabled.
The damage is easy to miss. site_name has a built-in fallback, so
getStringOrDefault renders the cleared value as the default product name and the
login page visibly changes, while the toggles just go quiet. Reopening the
settings page reads the already-cleared state back into the form, so correcting
the one visible field and saving persists the rest of the damage.
Fields that grew their own guard already survive this: the SMTP block falls back
to the previous values when smtp_host arrives empty, secret fields are written
only when non-empty, and 132 request fields are pointers whose handler merges an
omitted field with the stored value. This generalizes that pattern rather than
adding a fourth ad-hoc guard.
The handler now decodes the payload a second time as a raw field map, resolves
the setting key each absent field would have written, and hands that set to the
service, which drops those keys before SetMultiple, so the stored value is never
touched. Fields the payload does carry are written as before, giving the caller
the partial-update semantics it was already assuming. The mapping is reflected off
the request's json tags so new fields are covered without maintaining a list;
smtp_from_email is the only field whose json name differs from its setting key
and is aliased explicitly.
Only value-typed fields are filtered. Pointer fields keep whole-document
behaviour on purpose: forwarded_client_ip_headers and
api_key_acl_trust_forwarded_ip depend on being rewritten on every save to
re-normalize fail-closed state, which the malformed forwarded-client-IP header
test pins down.
An explicitly sent empty value is still a deliberate clear; only absent fields
are preserved. A partial write refreshes the in-process caches from storage
instead of from the request struct, which holds zero values for whatever the
caller omitted.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Registration duplicate checks compared the email verbatim (lowercase +
trim only), so a single inbox could spawn unlimited accounts via provider
alias features: plus addressing (user+tag@gmail.com) and the Gmail dot
trick (u.s.e.r@gmail.com) both deliver to the same mailbox but were seen
as distinct, verifiable addresses. This lets abusers bulk-register to farm
signup grants while the domain whitelist and email verification pass.
Add NormalizeEmailForAliasDedup to collapse these variants to a single
"inbox identity" and use it (via existsByEmailOrAlias) on the three local
email-registration paths: register, send-verify-code, and its async
variant. The exact ExistsByEmail check runs first; only on a miss do we
scan the candidate domains (gmail.com/googlemail.com are mutual aliases)
and compare normalized forms. Lookup errors fail closed, matching the
existing check, so the path cannot be bypassed by inducing errors.
Scoped to registration only — email storage, display, login, and delivery
are unchanged. OAuth-bound emails (already provider-verified) are out of
scope.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Supersedes the earlier attempt #4439 by @KnWsdHunter, which its own author closed on 2026-07-20 and which was never merged. This PR discloses that prior work in its description and contributes a broader 6-case test table. Courtesy credit to @KnWsdHunter for the original diagnosis.
Composite model routes support a match_type: prefix mode with an optional
upstream_model. Leaving upstream_model empty is meant to pass the concrete
requested model through to the upstream, but every request matching the prefix
collapsed to the route's public_model instead. A route with
public_model=deepseek-v4, match_type=prefix and an empty upstream_model
forwarded both deepseek-v4-flash and deepseek-v4-pro upstream as deepseek-v4.
Root cause: normalizeCompositeRouteInput backfilled an empty upstream_model with
public_model at save time, so CompositeRouteResolver.Resolve never reached its
"upstreamModel == '' -> concrete requested model" fallback.
- Skip the backfill for prefix routes only; an empty upstream_model stays empty
and Resolve passes the concrete requested model through.
- exact routes keep the backfill (model == public_model, so persisted/display
contract is unchanged).
- Explicit upstream_model still forwards that fixed model in every mode.
- Frontend: add an upstream_model hint (zh/en) documenting empty = passthrough.
Avoid recording the generic account-model transient cooldown for pool-mode statuses explicitly configured for same-account retry. This lets the bounded request-local retry budget complete while preserving cooldowns for other statuses and non-pool accounts.
Composite public aliases (e.g. all/claude) reach the Anthropic/Gemini
billing core via OriginalModel/ChannelMappedModel source overrides.
Unknown aliases resolved to no pricing and silently recorded $0 cost,
while family-word aliases were mispriced by the fallback family match
(Opus traffic billed at the Sonnet fallback rate). The OpenAI path
already guards this via usageBillingModelCandidates; the shared
recordUsageCore had neither the guard nor a fallback.
- composite groups: unless the admin explicitly configured channel
pricing for the alias (OpenRouter-style custom pricing), bill by the
concrete forwarded model
- general safety net: when the selected billing model has no resolvable
pricing at all, fall back to the concrete forwarded model instead of
silently recording $0
- grok media usage records now attribute OriginalModel to the client
requested public alias, consistent with every other endpoint
(billing unaffected: empty BillingModelSource never triggers source
overrides)
Priced traffic and non-composite groups are unaffected.
The composite-groups PR (#3581) replaced the raw ent SubscriptionPlan
response of GET /admin/payment/plans with a projection struct but
dropped the currency field added by #4323. PlanEditDialog then read an
undefined currency, sent an empty string on save, and silently wiped
the stored plan currency. Restore currency plus created_at/updated_at
so the projection preserves the full original response shape.
Replace the broad text response matcher with an explicit non-SSE MIME allowlist. Document proxy behavior and enforce the canonical Caddy compression policy in CI.
formatMultiplier rounded any value >= 0.01 with toFixed(2), so a
configured multiplier like 0.035 displayed as 0.04x in the /usage and
/admin/usage cost tooltips. Format with up to 4 decimals and trim
trailing zeros while keeping at least 2 decimals; display-only fix,
billing amounts were already computed from the true multiplier.
govulncheck newly flags GO-2026-5970 (infinite loop on invalid input in
golang.org/x/text < v0.39.0), causing every fresh backend-security CI run
to fail. Upgrading x/text pulls sibling golang.org/x/* modules
(crypto/mod/net/sync/sys/term/tools) to their matching minor versions.
Remove unsupported none handling, align the auth cache version with the current baseline, consolidate the unshipped migrations, and update the public API contract.
Persist reasoning ceilings and exact mappings for OpenAI groups, enforce them across HTTP and WebSocket forwarding, and invalidate cached auth snapshots.
Debounced requests can finish out of order, allowing an earlier response to replace newer results or repopulate a cleared field. Track a request generation and gate state updates, with deterministic regression coverage.
Constraint: Preserve the existing 300 ms debounce and admin usage API contract
Rejected: Add AbortController plumbing | sequence invalidation matches repository patterns without changing API signatures
Confidence: high
Scope-risk: narrow
Reversibility: clean
Directive: Keep async filter results gated against superseded queries
Tested: Focused Vitest (4 tests), vue-tsc, ESLint, and full frontend suite except two pre-existing rollback assertion failures
Not-tested: Live browser against production-sized user data
Related: https://github.com/Wei-Shaw/sub2api/issues/4551
Fixes#4653
## Problem
The /v1/responses endpoint (Responses → Anthropic bridge) failed when
upstream providers returned SSE events in compact format without spaces
after colons (e.g. 'event:message_start' instead of 'event: message_start').
This affected Kimi API (api.kimi.com) and potentially other upstreams,
causing 'Upstream stream ended without a response' errors.
## Root Cause
The SSE parser in gateway_forward_as_responses.go used strict prefix
matching with hardcoded spaces:
if !strings.HasPrefix(line, "event: ") { ... }
if !strings.HasPrefix(dataLine, "data: ") { ... }
According to the SSE spec
(https://html.spec.whatwg.org/multipage/server-sent-events.html),
the space after the colon is optional. Both formats are valid:
event: message_start ✓
event:message_start ✓
## Solution
Introduced parseAnthropicSSEField() helper that:
- Accepts both 'field: value' and 'field:value' formats
- Uses TrimSpace() to handle optional/multiple spaces
- Returns (value, ok) for consistent error handling
Applied to both streaming handlers:
- handleResponsesBufferedStreamingResponse
- handleResponsesStreamingResponse
## Testing
Added comprehensive unit tests:
- TestParseAnthropicSSEField: validates parser with 8 edge cases
- TestHandleResponsesBufferedStreamingResponse_CompactSSEFormat
- TestHandleResponsesStreamingResponse_CompactSSEFormat
All existing tests pass, confirming backward compatibility.
Resolve const-block conflict in openai_gateway_grok_cache.go:
keep #4590's client tool cache constants, drop grokFreeRolling24hTokenLimit
(moved to pkg/xai as IsGrokFreeRolling24hTokenLimit with legacy 2M support).
Subscription plans can carry a display-only ISO currency code, but the admin plan table and user plan cards still used a hard-coded dollar sign. Resolve the symbol through the shared currency helper while retaining existing number formatting, the currency code label, and legacy blank-currency behavior.
Constraint: Plans without a currency must keep the legacy USD display
Rejected: Reformat complete prices with Intl.NumberFormat | it would change existing decimals and remove the explicit currency code
Confidence: high
Scope-risk: narrow
Reversibility: clean
Directive: Keep plan currency symbols aligned across admin and user displays
Tested: targeted Vitest 4 tests; targeted ESLint; vue-tsc; Vite production build; full Vitest 1222 tests passed
Not-tested: Full Vitest has 2 unrelated upstream failures in admin.system.rollback.spec.ts
Related: https://github.com/Wei-Shaw/sub2api/issues/4609
Grok OAuth 'test connection' reused the production GrokTokenProvider
request path, whose eligibility gate rejects any account that is not
currently schedulable (manual switch off, rate limited, overloaded,
temporarily cooled down). Every such test failed before reaching xAI
with the generic 'oauth refresh account state changed', while
Codex/OpenAI account tests read credentials regardless of scheduling
state (#4598).
Add GetAccessTokenForManualTest and use it from the account test
service:
- skips the request-path scheduling eligibility gate so admins can
probe excluded accounts; credential integrity checks (configured
proxy present, refresh token present) still apply
- refreshes expired tokens through the shared OAuthRefreshAPI without
the request-path marker, so the lock protocol and DB re-read keep
working while scheduling state no longer blocks the refresh;
credential rotation for non-active accounts stays blocked inside
RefreshIfNeeded
- falls back to the still-valid current token when a refresh attempt
fails or another worker holds the refresh lock, and otherwise
surfaces the concrete refresh error instead of the generic message
Fixes#4598
The in-place update ran entirely inside c.Request.Context(). Browsers and
reverse proxies commonly abort long-idle requests (axios global timeout
30s, nginx proxy_read_timeout 60s by default), which canceled the request
context mid-download and killed every slow update with
'download failed: context canceled' while the version stayed unchanged.
Users behind slow GitHub links saw the update button fail at a wall-clock
ceiling (~60s) on every attempt (#4504).
- Run PerformUpdate and RollbackToVersion on a context detached from the
request (context.WithoutCancel) and bounded by a 15-minute deadline so
the 10-minute GitHub download client owns its own timeout. A client
disconnect no longer aborts the binary swap; retries then hit the
system operation lock or report 'Already up to date'.
- Raise the frontend timeout for the update/rollback calls from the
global 30s axios default to 15 minutes so the browser can actually
wait for the result.
Fixes#4504
The Codex models manifest path returned upstream 401s straight to the
client without feeding them into the account state machinery. A revoked
or invalidated OAuth account therefore stayed active and schedulable,
kept being selected for subsequent /models requests, and produced
repeated 502s until an admin ran a manual connection test (#4544).
- Route ChatGPT-backend manifest 401s through the shared upstream-error
handling: token cache invalidation, temp-unschedulable cooldown for
refreshable OAuth accounts, permanent disable for
token_revoked/token_invalidated, plus the runtime scheduling block.
- Treat ChatGPT-backend manifest 401s as failover-eligible so the
current /models request can switch to a healthy account instead of
returning 502. Custom API key upstream 401s keep the existing
no-failover, no-disable behavior since their /models auth is not
authoritative for the account.
- Skip Agent Identity accounts: their 401s can be task-scoped and have
a dedicated recovery flow.
- Attach the selected account to the ops error-log context so /models
failures record the account_id.
Fixes#4544
Co-authored-by: Cursor <cursoragent@cursor.com>
Root cause: Chat bridge system messages were mirrored to instructions while remaining in input as developer messages.
Fix: omit only losslessly promoted system messages for non-JSON-object OAuth Chat requests while preserving compatibility paths.
Closes#4603
Route first-turn pre-response transport, HTTP, and SSE failures through the existing account failover path without replaying later turns. Require a real Responses terminal event and preserve platform-specific OpenAI/Grok error side effects.
ProviderSet 提供了 NewPromptAdminHandler，也绑定了 PromptEngine，却没有为
PromptAdminService 绑定实现，导致 go generate ./cmd/server 直接失败：
no provider found for securityaudit.PromptAdminService
needed by *securityaudit.PromptAdminHandler
*PromptService 本就实现了该接口的全部方法，wire_gen.go 也一直是手工维护到位的
（补上绑定后重新生成的结果与现有文件逐字节一致），只是生成这条路被堵死了，
后续任何依赖改动都无法再靠 go generate 同步。
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01VHreE5pzCkSYz7J45fmd2Y
Round-trip Grok encrypted reasoning via Anthropic thinking.signature and
seed cache identity from Claude Code session IDs, so multi-turn /v1/messages
traffic can hit xAI prompt cache more like CLIProxyAPI. Also strip invalid
encrypted_content and retry on the Messages path when decrypt fails after
account switches.
deploy/.env.example documents POSTGRES_MAX_CONNECTIONS,
POSTGRES_SHARED_BUFFERS, POSTGRES_EFFECTIVE_CACHE_SIZE and
POSTGRES_MAINTENANCE_WORK_MEM, with notes on how to size them — but no
compose file ever passes them to the postgres container. A user who sets
them in .env gets nothing, silently.
Wire them into the postgres command in deploy/docker-compose.yml. The
fallbacks are the postgres:18 stock defaults (100 / 128MB / 4GB / 64MB),
so a deploy that does not set the variables behaves exactly as before.
Checked with postgres:18-alpine: with the variables unset, SHOW gives
the stock values; with them set (1024 / 1GB / 6GB / 128MB), SHOW gives
the set values.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
PR #4506 fixed this in deploy/docker-compose.yml, but the same broken
form is still in docker-compose.dev.yml and docker-compose.local.yml.
The redis command is one quoted script given to the inner sh -c, and
compose keeps the newlines inside the quoted string, so redis-server on
the first line runs as a complete command with no flags at all. The
--save / --appendonly / --appendfsync lines are silently never applied,
and ${REDIS_PASSWORD:+--requirepass ...} is dead too — redis takes no
password even when REDIS_PASSWORD is set.
The fix is the same trailing `\` line continuations as #4506, with the
same comment, so the three compose files read the same way.
Checked with both files on redis:8-alpine, REDIS_PASSWORD set. Before:
PING with no auth said PONG, appendonly was "no", save was the stock
"3600 1 300 100 60 10000". After: no-auth PING gets NOAUTH, appendonly
is "yes", save is "60 1". With REDIS_PASSWORD unset the server still
starts open, as before.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Use *string for AnthropicResponse.StopReason so streaming message_start
serializes as JSON null instead of "", matching official Anthropic wire
format and avoiding strict client turn-finalization failures.
BatchImageGuideView carried ~140 user-visible strings outside vue-i18n:
hardcoded Chinese template strings (buttons, placeholders, tooltips,
table headers, dialog titles, empty states) plus an inline bilingual
batchImageText dictionary switched by isZhLocale() in script.
Move all of it into a new batchImage locale namespace (180 leaf keys
per language, zh/en verified key-aligned) registered in both locale
indexes, reusing 15 existing common.*/pagination.* keys. Column and
filter-option definitions become computed so language switches take
effect immediately; countable strings use vue-i18n plurals and the
three rich-text counters keep their inner bold spans via <i18n-t>
slots.
The Codex agent-instruction prompt template intentionally stays as a
plain string: it is an AI-facing prompt document containing literal
JSON braces that conflict with vue-i18n message syntax.
Verified: eslint, vue-tsc --noEmit, vite build, and the full vitest
suite (175 files / 1219 tests) all pass.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Move the last few user-visible hardcoded strings into vue-i18n:
- AppHeader: "Toggle Menu" / "User Menu" aria-labels (screen readers
read these in English regardless of locale)
- CustomPageView: "Page not found" error body
- PaymentProviderDialog: display-name example placeholder hardcoded in
Chinese for all locales
Adds common.toggleMenu / common.userMenu / common.pageNotFound and
admin.settings.payment.customMethodDisplayNamePlaceholder to both zh
and en locales, and registers the new placeholder key in the
PaymentProviderDialog spec's i18n mock.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
The project's dark theme uses the custom slate-based dark-* palette
(928 dark:bg-dark-* / 787 dark:border-dark-* occurrences), but a
minority of surfaces still used Tailwind's neutral gray-* scale, which
has a visibly cooler hue next to slate cards — most noticeably the
filter dropdown panels on the usage, accounts and subscriptions pages.
Migrate the remaining dark:bg-gray-* / dark:border-gray-* (and their
hover: variants) to the dark-* palette across 12 components: filter
dropdowns, chart card borders, menu separators, and small badges.
49 class-string substitutions, no structural or logic changes;
dark:text-gray-* is untouched since gray text on dark surfaces is the
established project-wide convention.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
The user info banner in UserBalanceModal renders on a dark:bg-dark-700
container, but the email line kept text-gray-900 and the balance line
text-gray-500 with no dark variants, leaving near-black text on a dark
background. Align both with the pattern already used elsewhere in the
same dialog (text-gray-900 dark:text-gray-100 / text-gray-500
dark:text-gray-400).
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
The available channels table rendered inside its own
"card overflow-hidden" wrapper instead of the ".table-wrapper" element
that TablePageLayout's scroll chain expects. The outer
.table-scroll-container is overflow-hidden, so when the channel list
grew taller than the viewport the content was simply clipped with no
scrollbar. The extra card wrapper also drew a double border inside the
layout's own card chrome.
Mount the table on .table-wrapper like every other table page so it
receives overflow-y-auto from the layout, and add a source-convention
regression test in the style of TablePageLayout.spec.ts.
Fixes#4555
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Custom API key upstreams that only implement the standard OpenAI
GET /v1/models endpoint return {"object":"list","data":[{"id":...}]},
which the Codex models manifest validation rejected with
"missing top-level models array", surfacing HTTP 502 to Codex clients.
Convert the standard list envelope into the Codex manifest shape
({"models":[{"slug":...}]}) for API key upstreams before validation.
Bodies that already carry a top-level models field, are not the
standard list shape, or yield no usable model IDs are left unchanged
so existing validation and account failover behavior is preserved.
Fixes#4490
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Reconcile with #4539 (grok media account model mapping), now on main:
- handler/grok_media.go non-failover error path keeps both changes —
#4539's grokMediaScheduleModel(account, routingModel, nil) schedule
attribution and this branch's IsResponseCommitted guard
- auto-merged sections verified: routing/classify use #4539's routingModel,
video lookup owner-binding and no-failover semantics intact, ForwardGrokMedia
keeps mapping block (skipped for lookup endpoints via RequiresRequestBody),
empty-image failover, and video-status URL rewrite in order
Resolve conflicts with main:
- service/grok_media.go: keep both post-response blocks — #4497's empty
image-output failover (main) runs first for image endpoints, then this
branch's video-status content-URL rewrite; the endpoint conditions are
mutually exclusive
- handler/openai_gateway_credential_failover_loop_test.go: mark the stub
OAuth accounts media-eligible via the grok_media_eligible extra override,
because this branch moved grok media failover coverage to the generation
endpoint, which is now gated by #4540's paid-eligibility probe on main
Resolve conflicts with main:
- handler/grok_media_test.go: keep both new tests (schedule model test from
this branch, eligibility gating tests from #4540)
- service/openai_gateway_grok_test.go: keep both new tests; update the image
cases of the mapping table test to return a non-empty image payload because
#4497 (already on main) now converts empty image responses into an upstream
failover error
Building the image for linux/amd64 on an arm64 host (Apple Silicon)
kept failing at 'go mod download' with 'unexpected EOF'.
Two changes:
1. The builder stages had no --platform, so the Node and Go toolchains
ran under QEMU emulation of the target arch. Pin frontend-builder
and backend-builder to $BUILDPLATFORM and cross-compile: add
TARGETOS/TARGETARCH args and set them on 'go build'. The binary is
CGO_ENABLED=0, so this is a clean pure-Go cross-compile - much
faster, and the emulated networking that dropped module fetches
with EOF is gone. The frontend output is JS (arch-neutral), so it
is safe to build on the host arch too.
2. Add go module and build cache mounts, so a retry after a network
blip goes on from where it stopped instead of starting over.
The runtime image and app behavior are unchanged.
The redis command is one quoted script given to the inner `sh -c`.
Docker compose keeps the newlines inside the quoted string, so
`redis-server` on the first line ran as a complete command with no
flags at all, and --save / --appendonly / --appendfsync after it were
silently never applied (`redis-cli CONFIG GET appendonly` said "no").
Trailing `\` line continuations fold the script back into one command.
Checked with redis:7-alpine: appendonly is now "yes" and save is
"60 1".
- Add prompt_audit_events.full_prompt (migration 182) so admins can review
the exact unredacted prompt that triggered a finding; blocking mode writes
it from the snapshot, async mode reconstructs it from the Redis scan
payload so jobs rows stay redaction-only
- Event detail API returns full_prompt (list endpoint stays lean); text is
NUL-stripped and capped at 65536 runes
- Detail dialog shows the full prompt in a scrollable pane with fallback to
the legacy redacted preview; page copy updated to match the new behavior
- Rework filter deletion into a dedicated dialog with time-range presets and
criteria-change preview invalidation; localize decision/risk/category
labels across the events workspace
- Fix pre-existing i18n message-compile spec by declaring the
@intlify/message-compiler dev dependency
Let admins configure private/intranet Guard endpoints without destination-class blocking, and fix prompt-audit switch layout so thumbs and labels no longer overlap.
Co-authored-by: Cursor <cursoragent@cursor.com>
The Anthropic→Responses streaming converter omits two things the OpenAI
Responses wire format requires, which breaks clients that reconstruct the
response from the event stream (rather than just reading deltas).
1. response.content_part.added is never emitted.
A message item is opened with content: [], and the OpenAI SDK's
accumulating stream helper (client.responses.stream) only appends a
content part when it sees content_part.added. Without it, the next
output_text.delta indexes output.content[content_index] and raises:
File "openai/lib/streaming/responses/_responses.py", line 352,
in accumulate_event
content = output.content[event.content_index]
IndexError: list index out of range
Raw iteration (responses.create(stream=True)) does not accumulate and is
unaffected, which is why this went unnoticed.
2. response.completed carries Output: []ResponsesOutput{}.
get_final_response() and tracing integrations parse the terminal event's
response directly, so callers see an empty output_text even though the
deltas streamed correctly. This one is invisible when only watching the
stream render.
Also carries the full text on output_text.done / content_part.done (deltas
carry increments only, done events carry the whole part) and fills in
content/arguments/summary on output_item.done, which had the same
empty-payload issue.
Reproduced against a live Anthropic-platform group with the openai Python
SDK 2.46.0; Arize Phoenix's playground hits the same path. Verified before
(IndexError) and after (full text via get_final_response()).
Adds regression tests covering event ordering, done-event payloads, and the
terminal event's output for both text and tool calls.
Behind a reverse proxy (e.g. nginx with X-Real-IP), admin audit logs and
session IP/UA binding always recorded 127.0.0.1 because they hardcoded
the gin trusted_proxies chain, while API key IP restriction already
honored the "trust forwarded client IP" system setting.
- add ip.GetSecurityClientIP(c, trustForwarded) as the single source of
truth for security-sensitive client IP selection; API key auth
middlewares (main + google) refactored onto it with zero behavior change
- SessionBindingContext(cfg) now resolves the client IP via the same
toggle and injects it into the request context; token issuance,
binding enforcement and its mismatch audit record all read the
injected value, so issue/verify can never diverge
- audit log middleware and audit-log clear trace record the same
security client IP (middleware.SecurityClientIP), falling back to the
trusted proxy chain when the injection is absent
- settings UI hint (zh/en) documents the broadened toggle scope and the
one-time re-login after toggling while session binding is enabled
With the toggle off (default) behavior is byte-for-byte unchanged.
Co-Authored-By: Claude <noreply@anthropic.com>
- admin.accounts.oauth.openai.mobileRefreshTokenAuth was referenced by
OAuthAuthorizationFlow.vue since 9f8cffe88 (Mobile RT entry) but never
added to zh/en locales, rendering the raw key in the add-account wizard
- admin.accounts.oauth.openai.accessTokenAuth has the same latent issue
since 26060e702 (Sora AT import); currently hidden but fixed alongside
Co-Authored-By: Claude <noreply@anthropic.com>
Scan client-injected assistant/tool/model turns, fail closed when config cannot
be trusted after startup or stale invalidation, reuse probe tokens only for the
same base URL, and restrict localhost dials to loopback addresses.
Co-authored-by: Cursor <cursoragent@cursor.com>
Reuse applyGrokFreeMessagesFunctionToolCacheRoute on native /v1/responses
and the Grok WS HTTP bridge so Free OAuth requests with client function
tools get the same mixed-tools cache route as the Messages bridge
(append/convert web_search and x_search).
Also dedupe: Grok Build already declares function tools named web_search,
so naive append caused "Duplicate tool names: web_search". Convert those
function entries to native tool types and skip duplicates.
Only Free OAuth accounts (isKnownGrokFreeAccount); paid/unknown unchanged.
PR #4425 was authored before #4429 widened NewUserHandler with the
step-up TOTP and user services, and merged without a rebase, breaking
typecheck on main.
Reconcile the OAuth media route with the manual endpoint-switch redesign
(7f5d067af): media leaves for api.x.ai only when text traffic resolves to
the CLI gateway host; manually selected official/regional/custom endpoints
keep serving media as-is.
Admins often recreate groups with the same pricing, routing, and account membership. A server-side duplicate creates an inactive copy for review, preserves eligible account priorities, and recovers ambiguous retries without creating extra groups.
Constraint: Group has no neutral JSON metadata field for durable operation recovery
Constraint: Model routing references account IDs, so copied configuration requires matching bindings
Rejected: Rebuild from the list response | it omits configuration and account priority details
Rejected: Store operation identity in business configuration | it would pollute real group settings
Confidence: high
Scope-risk: moderate
Reversibility: clean
Directive: Keep duplicated groups inactive until an administrator reviews the copied configuration
Tested: Go unit and full tests, go vet, integration-tag compile, frontend Vitest, lint, typecheck, production build, and Playwright duplicate flow
Not-tested: PostgreSQL container integration locally because Docker is unavailable; CI will execute the database-backed suite
Admins often recreate monitors with the same endpoint, model, and request settings. A server-side duplicate keeps the stored API key out of the browser, creates a disabled copy for review, and uses stable operation identity to recover ambiguous retries without creating extra rows.
Constraint: Stored monitor API keys must never be returned to the browser
Constraint: Applying a request template must preserve internal duplicate recovery metadata
Rejected: Rebuild the monitor from list data | list responses only contain a masked API key
Rejected: Copy runtime state and history | a duplicate should start as an unverified configuration
Confidence: high
Scope-risk: moderate
Reversibility: clean
Directive: Keep duplicated monitors disabled until an administrator reviews and enables them
Tested: Go unit tests for repository, service, and admin handler; integration-tag compile; go vet; golangci-lint v2.9; frontend Vitest, ESLint, typecheck, production build; Playwright duplicate flow
Not-tested: PostgreSQL container integration locally because Docker is unavailable; CI will execute the database-backed suite
For OpenAI-compatible API-key accounts, /v1/responses requests with
image-generation intent could be scheduled to accounts whose upstream
does not support the Responses API (extra.openai_responses_supported=false).
The flag was only consulted at forward time, where such accounts are
silently downgraded to a Chat-Completions path that cannot produce images,
causing upstream 4xx/5xx or canceled requests.
Fix:
- Add endpoint capability OpenAIEndpointCapabilityResponses. Its check in
SupportsOpenAIEndpointCapability excludes only OpenAI API-key accounts
probed as unsupported (mirroring the forward-time downgrade condition);
OAuth/Grok/unprobed accounts keep existing behavior, and a responses-
capable upstream must still pass the chat_completions gate. Reusing the
existing requiredCapability plumbing makes every scheduler filter path
enforce it with no scheduler signature changes.
- Request the responses capability at the HTTP Responses and
ResponsesWebSocket call sites only when imageIntent && platform==openai,
so non-image requests keep the downgrade path and Grok's own image path
is untouched.
- Normalize max_tokens -> max_output_tokens on the native responses
forward path (PlatformOpenAI), and strip prompt_cache_options alongside
prompt_cache_retention/safety_identifier.
/v1/images/generations continues to use native image capability (unchanged).
Tests: capability truth table, scheduler exclusion of unsupported accounts,
and forward-path transform behavior.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01KXpzKKvsb5jW2GvgBQqnnZ
The eligibility gate in grokChatResponsesBridgeEligibility rejected any
structured content (arrays), including standard OpenAI image_url parts.
This forced image-bearing requests to fall back to raw Chat Completions
forwarding, where non-Composer models have no image bridge — silently
dropping the images or causing upstream errors.
Changes:
- Allow content arrays containing only text and image_url parts to pass
eligibility, so ChatCompletionsToResponses can convert them to
input_image parts that Grok vision models natively support.
- Force image-bearing requests to use Responses even when no prompt
cache identity is available, since the raw path cannot forward images
for non-Composer models.
- Add grokChatStructuredContentBridgeable helper with conservative
validation: unknown part types (e.g. input_audio) still fall back.
- Update tests: image_url content is now bridgeable; add cases for
text+image, text-only arrays, unknown parts, and empty arrays.
Add an opt-in first semantic output budget for native HTTP Responses, including response-header wait. Keep preamble and keepalive bytes non-semantic so a stalled account can fail over once without replaying its response IDs. Defaults remain disabled.
Related to #4201, #4185, and #4248. Complements the HTTP/2 dead-connection fix in #4207.
Adversarial re-verification found one more divergence from the
double-conversion chain: an upstream 200 with empty choices (or a nil
response) left stop_reason as an empty string, while the old chain
reports end_turn. Derive the fallback from the content blocks — the
guard never fires when choices exist, since every finish_reason maps to
a non-empty stop_reason.
Also strengthen the equivalence tests: compare tool_use Content[].Input
and tool_call Function.Arguments across bridges, and add an
empty-choices stop_reason parity case.
Adversarial review of the direct bridge found divergences from the
double-conversion chain it replaces; all are now aligned and covered by
tests that fail against the previous implementation:
- flush argument fragments buffered before a deferred tool announcement,
and announce name-less tools at finalize, so no tool arguments are lost
when upstreams stream arguments before the name
- fold text-only user array content into a single string; parts form only
when an image requires it (strict chat upstreams reject array content)
- drop tool_choice pointing at undeclared tools and unknown choice types
- treat cache_write_tokens/cache_creation_tokens as alternate spellings
(prefer write), not additive
- generate a response id when the upstream omits one
- derive stop_reason from blocks for content_filter/unknown finish reasons
- emit input_json_delta "{}" when a tool block closes without argument
deltas
Ambiguous idempotency-store failures can occur after the account transaction commits. Scope durable recovery markers to the authenticated admin, retain the operation key across reloads, and recover only an already committed copy without rerunning active work.
Constraint: Generic idempotent handlers may legitimately remain active while a recovery lookup is attempted
Rejected: Reclaim or rerun an in-progress duplicate request | can execute account creation concurrently
Rejected: Recover by source account and key alone | allows another admin to observe the committed copy
Confidence: high
Scope-risk: narrow
Reversibility: clean
Directive: Keep ambiguous-response recovery read-only and bind durable operation markers to the authenticated actor
Tested: Full Go unit suite, go vet, server build, integration-test compilation; frontend lint, typecheck, 1,030 Vitest tests, and production build
Not-tested: Docker-backed PostgreSQL integration runtime because Docker is unavailable
Related: Wei-Shaw/sub2api#1379
Related: Wei-Shaw/sub2api#2928
Admins often need another account with the same provider and routing configuration. Duplicate on the server so credentials never return to the browser, preserve exact group priorities atomically, start the copy paused, and recover the same copy after ambiguous idempotency-store failures.
Constraint: Admin account responses redact credentials, so duplication must remain server-side
Constraint: OAuth and setup-token credentials rotate and must not be shared across account rows
Rejected: Copy raw account JSON to the clipboard | exposes credentials outside the server
Rejected: Duplicate rotating credentials | account-scoped refresh locks can race token rotation
Confidence: high
Scope-risk: moderate
Reversibility: clean
Directive: Keep copies paused, avoid automatic upstream probes, and exclude rotating credential types unless token ownership is redesigned
Tested: Targeted Go tests, Go vet, server build; frontend lint, typecheck, Vitest suite, production build; integration test compiled
Not-tested: Docker-backed PostgreSQL execution because Docker is unavailable
Related: Wei-Shaw/sub2api#1379
Related: Wei-Shaw/sub2api#2928
Three tests covering the CC→Responses→Anthropic finalize path:
- TestStreamingParallelToolUseNoGhostDelta: full end-to-end with two
parallel tool_calls, asserts every content_block_delta targets a
started block
- TestStreamingParallelToolUseSecondToolPackedArgsDone: focused test
where tool 1 streams deltas and tool 2 has packed .done args
- TestStreamingThreeParallelToolsAllPackedDone: three tools all with
packed .done, the most extreme case
All three fail on unfixed code (ghost deltas on wrong indices) and
pass after the fix.
Skip the Responses API intermediate representation on the /v1/messages
force-chat path. Previously every streaming token ran through two state
machines (CC→Responses→Anthropic); now it runs through one (CC→Anthropic).
New file backend/internal/pkg/apicompat/chatcompletions_anthropic_bridge.go:
- AnthropicToChatCompletionsRequest: request-side direct conversion
- ChatCompletionsResponseToAnthropic: non-stream response direct conversion
- ChatCompletionsChunkToAnthropicEvents + Finalize: single streaming state machine
openai_gateway_messages_chat_fallback.go rewired to use the direct bridge.
Existing 7 ForceChatCompletions end-to-end tests pass unchanged; 22 new
unit tests added including equivalence tests vs the double-conversion path.
resToAnthHandleFuncArgsDone used state.ContentBlockIndex directly
instead of looking up from OutputIndexToBlockIdx like
resToAnthHandleFuncArgsDelta does. When multiple tool_calls arrive
in parallel and arguments come as a packed .done (no prior delta),
the second+ tool would emit content_block_delta on an index that
was never content_block_start'ed, causing Claude Code to report
"Content block not found".
Fix: resolve block index from OutputIndexToBlockIdx, and skip the
delta if the block is already closed or the index doesn't match
the current open block.
Closes#4193
Mirror the Admin UI Server-Timing opt-in for user-facing pages so
authenticated callers can inspect total/app/db/redis/deps metrics on
session, profile, keys, usage, payment, and related user APIs.
- Collect when X-User-UI-Request=1 or path is on the user allowlist
- Emit for non-admin only on allowlisted paths (header is not auth)
- Exclude payment public/webhook surfaces
- Mark matching SPA requests and allow the new CORS request header
Add a dedicated client first-message timeout while preserving the legacy 30-second default.
Use the resolved value for both the WebSocket read deadline and structured timeout logs, and document tuning for large requests or slow links.
Add configuration, validation, handler, and resolver regression coverage.
Refs #4158
ChatGPT OAuth may normalize explicit image size controls to auto while returning smaller raster bytes. Decode final image headers so Images API metadata and tier billing reflect the actual output for both streaming and non-streaming OAuth responses.
Enrich openai.upstream_failover_switching structured logs with proxy
identity fields (proxy_id, proxy_name, proxy_host, proxy_port) so
operators can see which outbound proxy failed when switching accounts.
Matches the existing gateway.forward_failed / websocket_proxy_failed
logging pattern.
Add proxy_id/proxy_name/proxy_host/proxy_port fields to openai.websocket_proxy_failed so operators can identify which proxy failed without extra lookups. Matches the existing gateway.forward_failed logging pattern.
Add a toggleable group ID column in /admin/groups column settings.
It shows the group id (e.g. #1), is hidden by default for new and
existing admins via column-settings version migration, and remains
sortable against the existing backend sort_by=id support.
Expose API key ID in the keys table column settings so users can show
or hide it. Hidden by default; bump column-settings version so existing
preferences also keep ID hidden until toggled on.
WSv2 egress relays upstream events verbatim without the HTTP-path
namespace restore, so flattening requests that take the WSv2 branch
would surface flattened tool names the client cannot match. Resolve
the WS transport decision before flattening and skip flattening only
when the request will actually go WSv2 (passthrough accounts return
via HTTP before the WSv2 branch and still flatten).
Also check all type assertions in responses_namespace_test.go to
satisfy golangci-lint errcheck.
Resolve conflict in openai_gateway_passthrough.go streaming path: keep
main's normalizeCompletedImageGenerationStatus normalization ahead of
this branch's namespace restore block, mirroring the established order
in openai_gateway_response_handling.go.
The snapshot taken before dispatching the OAuth response handler still used
raw c.Writer.Size(), while handleOpenAIImagesOAuthResponseError compares it
against the keepalive-adjusted size. From the second failover round on, the
previous round's heartbeat bytes made the raw snapshot positive while the
adjusted size stayed -1, so a retryable upstream error was misjudged as
'response already written': the failover was swallowed and the client got a
committed 200 with a whitespace-only body.
Snapshot now uses OpenAIImagesJSONKeepaliveAdjustedWrittenSize, matching the
comparison side. Regression test simulates a prior-round heartbeat before
ForwardImages and asserts the retryable error still converts to
UpstreamFailoverError (verified red without the fix, green with it).
Codex/OpenAI upstream switched from a WebSocket pool to HTTP/2. The
outbound H2 transport set neither ReadIdleTimeout nor ResponseHeaderTimeout
(the OpenAI profile forces ResponseHeaderTimeout=0), so a pooled H2
connection silently killed by a proxy/NAT becomes a "dead connection":
both ends believe it is alive and a request assigned to it hangs until the
OS TCP retransmit timeout (minutes) before the first byte — observed as an
8m37s TTFT with an eventual 200. Occasional (only when a request lands on a
dead pooled conn) and across all groups (shared OpenAI transport); worse on
larger idle pools.
Explicitly configure http2 on the openai_h2 transport and enable active
PING health checks (ReadIdleTimeout=15s, PingTimeout=15s) so dead
connections are detected and evicted at the source, instead of relying on
ResponseHeaderTimeout as an after-the-fact backstop. Scoped to the
openai_h2 path only; Claude/Gemini (default) and h1 modes are untouched.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Treat a non-empty messages dispatch model as a caller-approved fallback after account-level mapping, without reclassifying the requested Claude family.\n\nAdd Fable forwarding and endpoint boundary regression coverage.\n\nFixes #4177
OpenAI OAuth (ChatGPT) accounts deterministically reject plan-gated
models with 400 "The 'X' model is not supported when using Codex with
a ChatGPT account". Account selection had no capability filtering for
this, so the scheduler kept picking the same account for the same model
forever; every attempt burned an upstream call and surfaced to clients
as a retryable 502, sustaining client retry storms.
Treat this 400 like upstream model-not-found: mark the (account, model)
pair via SetModelRateLimit (30min cooldown) so
IsSchedulableForModelWithContext skips the account for that model
during selection, and return true so the in-flight request fails over
to another account through the existing UpstreamFailoverError path.
<td>Thanks to APIKEY.FUN for sponsoring this project! <a href="https://apikey.fun/register?aff=SUB2API">APIKEY.FUN</a> is one of the core contributors to the sub2api open-source project, dedicated to providing open, stable, and cost-effective AI API access. The platform supports API relay services for Claude, OpenAI, Gemini, and other popular models, with pricing starting from as low as 7% of the original rate. Register via the exclusive link: <a href="https://apikey.fun/register?aff=SUB2API">APIKEY</a> to enjoy a permanent 5% discount on all recharges.</td>
<td>Thanks to APIKEY.FUN for sponsoring this project! <a href="https://apikey.fun/register?aff=SUB2API">APIKEY.FUN</a> is one of the core contributors to the sub2api open-source project, dedicated to providing open, stable, and cost-effective AI API access. The platform supports API relay services for Claude, OpenAI, Gemini, and other popular models, with pricing starting from as low as 7% of the original rate. Register via the exclusive link: <a href="https://apikey.fun/register?aff=SUB2API">APIKEY</a> to enjoy up to 5% off on all recharges.</td>
</tr>
<tr>
@@ -57,8 +59,8 @@ Please read the following carefully before using this project:
<td>Thanks to AICodeMirror for sponsoring this project! AICodeMirror provides official high-stability relay services for Claude Code / Codex / Gemini CLI, with enterprise-grade concurrency, fast invoicing, and 24/7 dedicated technical support. Claude Code / Codex / Gemini official channels at 38% / 2% / 9% of original price, with extra discounts on top-ups! AICodeMirror offers special benefits for sub2api users: register via <a href="https://www.aicodemirror.com/register?invitecode=KMVZQM">this link</a> to enjoy 20% off your first top-up, and enterprise customers can get up to 25% off!</td>
<td>Real GPT-5.6 series at 3% of OpenAI pricing — <a href="https://codex-everywhere.com">CodexEverywhere</a> is democratizing access to frontier models for developers worldwide. We believe in transparency and honesty, with model quality verified by active community oversight for months. USD and crypto friendly. Start with a free $20 trial at <a href="https://codex-everywhere.com">codex-everywhere.com</a>.</td>
</tr>
<tr>
@@ -77,14 +79,8 @@ Please read the following carefully before using this project:
<td>Thanks to PPToken.org for sponsoring this project! <a href="https://api.pptoken.org/register?promo=SUB2API">PPToken.org</a> specializes in GPT model API relay services, supporting Codex, Claude Code, OpenAI-compatible clients, and Gemini CLI integration. Top-ups are 1:1 (¥1 = $1 credit); GPT models start at 0.16x rate multiplier, with overall cost at roughly 2.2% of official pricing and first-token latency around 1 second — ideal for developers seeking low-cost, high-speed access to GPT model capabilities. Technical support: 24/7 real human responses (no bots), @tech in the group chat and get a reply within 10 minutes. Sponsor benefit: the first 200 users who register via the <a href="https://api.pptoken.org/register?promo=SUB2API">exclusive registration link</a> and enter promo code `SUB2API` can claim free Codex / Claude Code trial credits — no minimum spend, no card required.
<td>Thanks to Unity2 for sponsoring this project! <a href="https://unity2.ai/register?source=sub2api">Unity2</a> is a high-performance AI model API relay for individuals, teams, and enterprises, handling 30B+ tokens/day with 5000 RPM concurrency. One API Key works across Claude Code, Codex, OpenAI models, IDE plugins, and Agent workflows, with balance billing, bundled subscriptions, enterprise invoicing, and 1-on-1 support. <a href="https://unity2.ai/register?source=sub2api">Register</a> to claim $2 in balance, plus $10 more by joining the official group — up to $12 in free credit.
<td>Thanks to PPToken.cc for sponsoring this project! <a href="https://api.pptoken.cc/register?promo=SUB2API">PPToken.cc</a> specializes in GPT model API relay services, supporting Codex, Claude Code, OpenAI-compatible clients, and Gemini CLI integration. Top-ups are 1:1 (¥1 = $1 credit); GPT models start at 0.16x rate multiplier, with overall cost at roughly 2.2% of official pricing and first-token latency around 1 second — ideal for developers seeking low-cost, high-speed access to GPT model capabilities. Technical support: 24/7 real human responses (no bots), @tech in the group chat and get a reply within 10 minutes. Sponsor benefit: the first 200 users who register via the <a href="https://api.pptoken.cc/register?promo=SUB2API">exclusive registration link</a> and enter promo code `SUB2API` can claim free Codex / Claude Code trial credits — no minimum spend, no card required.
</td>
</tr>
@@ -100,55 +96,12 @@ Please read the following carefully before using this project:
<td>Thanks to Apikl for sponsoring this project! Built on Sub2API, the platform provides developers with relay services for Codex / Claude series models, focusing on long-term stability, high-speed direct connections, and excellent cost-effectiveness. It offers pay-as-you-go balance billing, enterprise-grade official invoices, and one-on-one dedicated support. <a href="https://apikl.ai">Register now</a> for a 1:1 top-up bonus — double your balance!
<td>Thanks to TokenEum for sponsoring this project! <a href="https://tokeneum.ai">TokenEum</a> is a comprehensive AI model aggregation platform and intelligent agent development company. It brings together top-tier international models — including Claude, Gemini, and OpenAI — alongside leading open-source models such as GLM, Qwen, and Kimi, offering a wide range of options across different quality and price tiers to suit every need. TokenEum also provides access to cutting-edge video generation models like Seedance2.0 and Happy Horse. Committed to transparency and honest business practices, TokenEum ensures all model information is accurate and reliable. Visit <a href="https://tokeneum.ai">tokeneum.ai</a> to get started.
<td>Thanks to Dola seed for sponsoring this project! Dola Seed 2.0 is a full‑modal general large model independently developed by ByteDance for the global market. Built on a unified multimodal architecture, it supports joint understanding and generation of text, images, audio, and video. It natively enables agent collaboration, with strong reasoning, long‑task execution, tool integration, and coding capabilities. It is widely applicable to smart cockpits, personal assistants, education, customer support, marketing, retail, and other scenarios. It excels in multimodal perception, end‑to‑end complex task delivery, stable interaction, and data security, and is readily accessible and deployable via the ModelArk platform.Register via <a href="https://dis.chatdesks.cn/chatdesk/hsyqsub2api.html">this link</a> to get 500,000 tokens of free inference quota per model.<a href="https://dis.chatdesks.cn/chatdesk/hsyqsub2api.html"> >>中国大陆地区的开发者请点击这里</a></td>
<td>Thanks to Suixiang AI Gateway for sponsoring this project! <a href="https://sui-xiang.com/">Suixiang AI Gateway</a> is a reliable and efficient API relay service provider offering relay services for Claude, Codex, Gemini, and more. A privacy-focused relay — no data reselling, no model dilution; privacy, transparency, and lightning-fast after-sales support. New accounts get ¥0.5 in trial credit daily by signing in; top-ups are 1:1, no subscription required, pay-as-you-go. Multi-line redundancy, cross-region disaster recovery, automatic failover, and uninterrupted long-link SSE. 99.9% availability — critical calls never fall behind.
<td>Thanks to MiyaIP for sponsoring this project! <a href="https://www.miyaip.com/?invitecode=sub2api">MiyaIP</a> is a platform dedicated to global residential proxy network services, committed to providing high-quality, pure overseas residential IP resources for enterprise developers, cross-border business teams, and AI application users. It delivers stable, independent overseas network environments for AI platforms, overseas SaaS, and other online services, supporting multi-region access testing and project environment isolation. Ideal for development and testing scenarios that require access to overseas AI services, such as: AI model platform access, AI development testing, AI SaaS service usage, AI API debugging, and multi-region network environment validation.
<td>Thanks to <a href="https://anpin.ai">anpin.ai</a> for sponsoring this project! anpin.ai is a premium AI relay service platform dedicated to advancing AI accessibility. With an advanced technical architecture and globally distributed deployment, it provides users with a direct high-speed channel to the world's top-tier large language models.<br>
Model authenticity: no content intervention or secondary filtering — experience the purest, most powerful native model capabilities.<br>
1:1 top-up, enterprise-grade service with invoicing available. Anpin AI is not just a relay — it's your secure, reliable, and efficient bridge to the frontier world of intelligence.
<td>Thanks to Proxy4Free for sponsoring this project! Proxy4Free is a data proxy service provider for developers and AI applications, offering residential proxies, static residential proxies, ISP proxies, and datacenter proxies for scenarios such as Web Scraping, Browser Automation, and AI Agents. With global IP resources, stable connections, and flexible switching, it helps developers improve data collection success rates and reduce the risk of IP bans. Register via <a href="https://www.proxy4free.com/?keyword=4yjqecpc">this link</a> to get started and easily build more stable and efficient automation workflows.
@@ -163,6 +116,62 @@ Model authenticity: no content intervention or secondary filtering — experienc
<td>Thanks to Aimzoon for sponsoring this project! <a href="http://aimzoon.com">Aimzoon</a> provides stable, cost-effective AI API access services, enabling developers to quickly connect popular AI services to coding tools such as Codex, Claude Code, and Gemini CLI. No complex configuration — faster onboarding, more stable calls, and lower costs. Ongoing promotions including discounted Codex rates and special pricing, with free trial credits upon registration, bringing AI coding into your daily workflow. <a href="http://aimzoon.com">Click here</a> to register and try it out!
<td><a href="https://nagora.ai/">Nagora</a> is a multi-model AI API gateway built for developers and teams. With a single account and API key, you can access more than 26 leading text and image models through one unified interface. It is compatible with OpenAI, Anthropic, and Gemini protocols and integrates seamlessly with development tools such as Claude Code, Codex, and Gemini CLI. The platform provides intelligent routing, automatic failover, transparent pricing, and consolidated billing, along with budget management, rate limiting, and concurrency controls. This makes AI usage more reliable and manageable across individual development, team collaboration, and production environments. No changes to your existing application are required. Simply replace the Base URL and API key to complete the integration in as little as one minute.</td>
<td>Thanks to <a href="https://www.novada.com/?sub2api/">Novada</a> for sponsoring this project! Novada provides residential, ISP, datacenter, and mobile proxies, along with Web Unlocker and Scraper APIs for developers building AI applications and automation workflows. With global IP coverage, flexible rotating and sticky sessions, and precise geo-targeting, Novada helps teams access web data reliably for AI agent workflows, cross-region testing, web research, and browser automation. Explore Novada to build more stable and scalable AI workflows.</td>
<td>Thanks to Qiniu AI for sponsoring this project! Qiniu AI is the enterprise-grade large-model MaaS platform under Qiniu Cloud (02567.HK), offering one-stop access to 150+ mainstream models worldwide, compatible with the protocols of major global model providers, and covering full-modality capabilities including text, image, audio, video, and file processing, serving over 1.69 million enterprises and developers. Qiniu AI offers an exclusive benefit for Sub2API users: register via <a href="https://s.qiniu.com/u6rQrq">this link</a> — enterprise users get 12 million tokens free, and developers get 3 million tokens free.</td>
<td>Thanks to FennoAI for sponsoring this project! FennoAI is a high-stability, high-performance API relay provider for enterprise R&D teams and developers, compatible with the OpenAI and Anthropic protocols and seamlessly integrating with mainstream AI coding tools such as Codex, Claude Code, and OpenCode. The platform delivers enterprise-grade stability, supporting call volumes of 100 billion tokens per day, and supports business-to-business settlement and invoicing for both domestic and overseas entities to meet enterprise R&D and procurement needs. As an exclusive benefit for Sub2API users, purchase a subscription via the <a href="https://api.fenno.ai/s/dC4k">exclusive link</a> to get $50 worth of Coding Plan credit for only $1.99. Referral rewards are also available: invite friends to purchase and earn up to 20% commission — the more you invite, the more you earn.</td>
<td>Thank you to LanoX AI for sponsoring this project! <a href="https://lanox.ai/?c=6">LanoX AI</a> provides stable, cost-effective global model access services for developers, teams, and enterprises. 🎁 New User Benefits — Claim millions of free tokens, plus 500+ free models for easier low-cost testing, validation, and deployment 🧠 Global Leading Models — GPT · Claude · Gemini · Qwen · Grok... 🎬 Multimodal Creation — Seedance 2.0 · GPT Image · Gemini Nano Banana 🛡️ Enterprise-Grade Reliability — High availability 💎 native capability output 💎 no intelligence degradation 💎 no model mixing 💎 transparent usage and billing 💎 💰 Lower API Costs — Top-tier models from as low as 10% of official pricing, with clear documentation, simple integration, invoicing support, and enterprise-scale batch usage 🏢 Enterprise Choice — Ideal for AI products, Agents, content platforms, and R&D teams with high-volume model usage</td>
<td><a href="https://www.rapidproxy.io/?ref=sub2api">RapidProxy</a> is a data collection proxy solution built for developers, providing stable and reliable residential proxy services. With 90M+ global residential IPs and 200+ country coverage, intelligent rotation, and precise geo-targeting, it helps projects such as web scraping, AI data training, SEO monitoring, and e-commerce data analysis break through access restrictions and improve data collection efficiency. It supports mainstream automation frameworks such as Playwright, Selenium, and Puppeteer, with prices as low as $0.65/GB — <a href="https://www.rapidproxy.io/?ref=sub2api">start your free test now</a>.</td>
<td><a href="https://hao.ai">hao.ai</a> is a high-speed, stable unified large-model API gateway for developers and teams. With a single API Key and a unified interface, you can access mainstream models such as GPT, Claude, and xAI Grok, with compatibility for common protocols and SDKs including OpenAI and Anthropic. The platform provides model routing, failover, team management, and complete request logs, with model prices as low as 15% of official reference pricing, helping users build AI applications more simply, more reliably, and at lower cost.</td>
<td>Swiftproxy is a high-performance proxy solution built for developers, providing stable and reliable residential and static residential proxy services. With 90M+ clean residential IPs, global coverage, flexible rotation, and precise geo-targeting, it helps projects such as web scraping, AI automation, browser automation, SEO monitoring, and multi-account management overcome access restrictions and improve workflow efficiency. It supports HTTP(S) and SOCKS5 protocols, integrates with popular automation tools like Playwright, Selenium, and Puppeteer, with dynamic proxy traffic that never expires until used and free testing available — <a href="https://www.swiftproxy.net/?ref=sub2api">start your free test now</a>!</td>
<td><a href="https://www.duckip.cn/?keyword=cu7oog6y">DuckIP</a> - 90M+ global residential network resources across 195+ countries and regions, with rotation and sticky sessions for public data collection, RAG updates, model evaluation, and multi-region data workloads. 🟢Residential Proxy - 20% Off; 🟢Static Residential Proxy - Starting at ¥50.00/IP; 🟢Unlimited Residential Proxy - Starting at ¥19.8/Hour. ✅Get 500M Free Trial.</td>
<td>Thanks to APIMart for sponsoring this project! <a href="https://go.apimart.ai/gh-sub2api">APIMart</a> is a low-cost API platform for AI image and video generation — GPT-Image-2 from $0.006 per image, with 160+ images per dollar. One async API covers both image and video: submit a task, get an ID, and retrieve results via polling or callback. Batch tens of thousands of images without timeouts, and switch models without changing code. Pay as you go with no monthly fee — <a href="https://go.apimart.ai/gh-sub2api">sign up here</a> to get started.</td>
</tr>
</table>
## Overview
@@ -179,6 +188,7 @@ Sub2API is an AI API gateway platform designed to distribute and manage API quot
- **Rate Limiting** - Configurable request and token rate limits
- **Built-in Payment System** - Supports EasyPay, Alipay, WeChat Pay, and Stripe for user self-service top-up, no separate payment service needed ([Configuration Guide](docs/PAYMENT.md))
- **Admin Dashboard** - Web interface for monitoring and management
- **Composite Groups** - Admin routing layer that resolves requested models to concrete providers for multi-provider groups ([Operator Guide](docs/COMPOSITE_GROUPS.md))
- **External System Integration** - Embed external systems (e.g. ticketing) via iframe to extend the admin dashboard
## Ecosystem
@@ -194,7 +204,7 @@ Community projects that extend or integrate with Sub2API:
| Component | Technology |
|-----------|------------|
| Backend | Go 1.25.7, Gin, Ent |
| Backend | Go 1.27.0, Gin, Ent |
| Frontend | Vue 3.4+, Vite 5+, TailwindCSS |
| Database | PostgreSQL 15+ |
| Cache/Queue | Redis 7+ |
@@ -524,6 +534,7 @@ database:
redis:
host:"localhost"
port:6379
username:""
password:""
jwt:
@@ -537,12 +548,6 @@ default:
rate_multiplier:1.0
```
### Sora Status (Temporarily Unavailable)
> ⚠️ Sora-related features are temporarily unavailable due to technical issues in upstream integration and media delivery.
> Please do not rely on Sora in production at this time.
> Existing `gateway.sora_*` configuration keys are reserved and may not take effect until these issues are resolved.
Additional security-related options are available in `config.yaml`:
-`cors.allowed_origins` for CORS allowlist
@@ -553,9 +558,18 @@ Additional security-related options are available in `config.yaml`:
-`security.csp` to control Content-Security-Policy headers
-`billing.circuit_breaker` to fail closed on billing errors
-`server.trusted_proxies` to enable X-Forwarded-For parsing
-`security.trust_forwarded_ip_for_api_key_acl` enables legacy raw forwarded-header takeover (enabled by default for upgrade compatibility); disable it to enforce `server.trusted_proxies`, which should contain only the exact proxy CIDRs that connect directly to Sub2API
-`security.forwarded_client_ip_headers` configures up to 16 third-party CDN client-IP header names; they are checked in order before the built-in headers only while legacy takeover is enabled
-`turnstile.required` to require Turnstile in release mode
Custom client-IP headers can be set in YAML or as a comma-separated environment variable:
Header names are validated, canonicalized, and de-duplicated. The admin security settings can update the list without a restart; new installations persist YAML/environment defaults and existing installations backfill a missing database value. When legacy takeover is disabled, all custom and built-in raw forwarding headers are ignored and Gin uses only `server.trusted_proxies`. While takeover is enabled, firewall the origin to CDN/proxy addresses and make the edge overwrite every trusted client-IP header. See [`deploy/EDGE_SECURITY.md`](deploy/EDGE_SECURITY.md) for the complete migration and trust-boundary rules.
**⚠️ Security Warning: HTTP URL Configuration**
When `security.url_allowlist.enabled=false`, the system performs minimal URL validation and **allows HTTP URLs by default** (dev-friendly mode; Docker Compose deployments use the same default). For production, explicitly tighten this to HTTPS-only:
@@ -605,17 +619,62 @@ per-turn user and account concurrency slots, which are released between turns.
```yaml
gateway:
openai_ws:
# Total time to receive and decompress the first client message.
client_first_message_timeout_seconds:30
# Close a client socket idle between completed turns; 0 disables this safeguard.
ingress_inter_turn_idle_timeout_seconds:300
# Distributed API-key limit for live client ingress sessions; 0 disables it.
max_ingress_connections_per_api_key:64
```
The first-message timeout is a total read deadline. Deployments that accept
large contexts or image-heavy requests over slower links can raise it to
120-300 seconds. It expires before HTTP bridge routing, so bridge mode does not
override this limit.
The connection cap is coordinated through Redis using a 60-second lease that
is refreshed every 20 seconds. A process that cannot confirm a lease for a
full lease lifetime closes its local WebSocket rather than continuing outside
the global cap. Use `http_bridge` for client-WebSocket/upstream-HTTP operation
when rolling out or mitigating upstream WebSocket issues.
the global cap.
Enable the v2 mode router before selecting an account-level WS mode such as
`http_bridge`:
```yaml
gateway:
openai_ws:
mode_router_v2_enabled:true
```
Or set `GATEWAY_OPENAI_WS_MODE_ROUTER_V2_ENABLED=true` in the environment.
Use `http_bridge` for client-WebSocket/upstream-HTTP operation when rolling out
or mitigating upstream WebSocket issues.
#### Force OpenAI upstream HTTP/SSE
When an egress proxy or network repeatedly reconnects OpenAI Responses
WebSockets, set the global fallback in the persisted deployment configuration:
```yaml
gateway:
openai_ws:
force_http:true
```
For Compose and Apple container deployments, the equivalent `.env` setting is:
```bash
GATEWAY_OPENAI_WS_FORCE_HTTP=true
```
This selects HTTP/SSE for OpenAI upstream Responses traffic that would
otherwise use WebSocket. It does not change the client-facing protocol or force
`GATEWAY_OPENAI_HTTP2_ENABLED=false`) separately when a proxy is incompatible
with HTTP/2. Unlike the account-level `http_bridge` mode, this global fallback
takes effect without enabling `mode_router_v2_enabled`. Keep the setting in the
deployment's persisted `.env` or `config.yaml`, rather than inside a running
container, so it is read again after an image update or container recreation.
#### ⚠️ Important: Creating the Admin Account
@@ -675,6 +734,12 @@ Simple Mode is designed for individual developers or internal teams who want qui
---
## Asynchronous Image Tasks
Long-running OpenAI/Grok image generation and editing can be submitted through `/v1/images/generations/async` or `/v1/images/edits/async`, then polled at `/v1/images/tasks/{task_id}` without holding a CDN connection open. See [Asynchronous Image Tasks](docs/ASYNC_IMAGE_TASKS.md) for request and response examples.
---
## Grok / xAI Support
Sub2API supports both Grok subscription accounts through xAI OAuth and standard xAI API-key accounts. Both account types forward OpenAI-compatible Responses traffic to xAI.
@@ -689,7 +754,8 @@ Sub2API supports both Grok subscription accounts through xAI OAuth and standard
- Codex CLI style Responses WebSocket ingress is accepted on the Responses targets and bridged to xAI HTTP/SSE Responses upstream
- Text models: `grok-4.5`, `grok-4.3`, `grok-build-0.1`, `grok-composer-2.5-fast`, `grok-4.20-0309-reasoning`, `grok-4.20-0309-non-reasoning`, and `grok-4.20-multi-agent-0309`
- Media targets for Grok groups: `/v1/images/generations`, `/images/generations`, `/v1/images/edits`, `/images/edits`, `/v1/videos/generations`, `/videos/generations`, `/v1/videos/edits`, `/videos/edits`, `/v1/videos/extensions`, `/videos/extensions`, `/v1/videos/{request_id}`, and `/videos/{request_id}`. Generation, editing, and extension requests require the group image-generation permission.
- Media models: `grok-imagine`, `grok-imagine-image-quality`, `grok-imagine-image`, `grok-imagine-edit`, `grok-imagine-video`, and `grok-imagine-video-1.5`
- Media models: `grok-imagine`, `grok-imagine-image-quality`, `grok-imagine-image`, `grok-imagine-image-2.0`, `grok-imagine-edit`, `grok-imagine-video`, and `grok-imagine-video-1.5`
- JSON image-edit and video-generation requests accept image references in `image`, `images`, `reference_images`, and `mask` objects. Use `url` for xAI-compatible payloads; the legacy `image_url` field remains accepted and is normalized to `url` before forwarding.
- Out of scope for this provider: TTS, transcription, browser automation, cookies, and Grok web scraping
### OAuth Configuration
@@ -704,7 +770,7 @@ The Grok OAuth flow uses PKCE and does not require committing private secrets. T
| `XAI_GROK_CLI_VERSION` | `0.2.93`; optional override for the client identity sent to `cli-chat-proxy.grok.com` |
| `XAI_GROK_CLI_VERSION` | `0.2.114`; optional override for the client identity sent to `cli-chat-proxy.grok.com`. The pinned value is also the floor: an override below it is dropped |
Administrators can create Grok OAuth or API-key accounts from the dashboard. OAuth authorization and reauthorization are also available through the admin API:
@@ -728,14 +794,13 @@ For API-key accounts, select **Grok → API Key** in the create-account dialog.
```toml
[models]
default = "sub2api-grok"
web_search = "sub2api-grok"
default = "grok"
web_search = "grok"
[model."sub2api-grok"]
[model."grok"]
model = "grok-4.5"
base_url = "https://your-sub2api.example.com/v1"
name = "Grok 4.5 via Sub2API"
description = "Grok 4.5 through a Sub2API Grok group"
name = "Grok 4.5"
api_key = "sk-your-sub2api-key"
api_backend = "responses"
context_window = 1000000
@@ -746,7 +811,7 @@ Back up an existing `config.toml` before merging the entry. The file contains a
```bash
grok inspect
grok -p "Reply with sub2api-ok" -m sub2api-grok
grok -p "Reply with sub2api-ok" -m grok
```
The `base_url` above is the public Sub2API URL ending in `/v1`, not `api.x.ai` or the internal xAI OAuth proxy URL.
@@ -757,6 +822,10 @@ xAI quota is passive. Sub2API does not invent subscription quota values; it reco
`401` responses temporarily remove accounts with invalid credentials from scheduling. `403` responses are treated as access or entitlement failures instead of token-refresh loops. `429` responses use `Retry-After` or a short cooldown to temporarily remove the account from scheduling.
New Grok image and video generation requests use a media-specific eligibility check. API-key accounts remain eligible. OAuth accounts require positive paid-entitlement evidence from the xAI billing probe; Free, forbidden, missing, malformed, and inconclusive billing observations are excluded from new media generation. Unobserved OAuth accounts are probed before the first media request is forwarded, and imports run the billing-first quota probe proactively. Chat requests and video status lookups are not affected by this media-only quarantine. If no eligible account remains, the media endpoint returns HTTP `503` with error type `grok_media_no_eligible_account`.
Administrators can override automatic media eligibility through the account create/update API by setting `extra.grok_media_eligible` to `false` (exclude) or `true` (force eligible). On update, set it to `null` to remove the override and return to automatic probe-based behavior; omitting the field preserves the current override. A weekly allowance period alone is not treated as a paid tier signal. Successful image responses must contain at least one actual image output; empty HTTP `200` responses trigger account failover instead of being counted and returned as successful generations.
<td>Real GPT-5.6 series at 3% of OpenAI pricing — <a href="https://codex-everywhere.com">CodexEverywhere</a> is democratizing access to frontier models for developers worldwide. We believe in transparency and honesty, with model quality verified by active community oversight for months. USD and crypto friendly. Start with a free $20 trial at <a href="https://codex-everywhere.com">codex-everywhere.com</a>.</td>
<td>感谢 MiyaIP 赞助本项目！<a href="https://www.miyaip.com/?invitecode=sub2api">MiyaIP</a> 是一家专注于全球住宅代理网络服务的平台，致力于为企业开发者、跨境业务团队及AI 应用用户提供高质量、纯净的海外住宅 IP 资源。为 AI 平台、海外 SaaS 及其他在线服务提供稳定、独立的海外网络环境，支持多地区访问测试和项目环境隔离。适用于需要访问海外 AI 服务的开发和测试场景，例如：AI 模型平台访问、AI 开发测试、AI SaaS 服务使用、AI API 调试、多地区网络环境验证
<td>感谢 Proxy4Free 赞助本项目！Proxy4Free 是面向开发者和 AI 应用的数据代理服务商，提供住宅代理、静态住宅代理、ISP 代理及数据中心代理等多种代理解决方案，适用于 Web Scraping、Browser Automation、AI Agent 等场景。支持全球 IP 资源、稳定连接与灵活切换，帮助开发者提升数据采集成功率，降低 IP 封禁风险。通过<a href="https://www.proxy4free.com/?keyword=4yjqecpc">此链接注册</a>即可开始体验，轻松构建更稳定、高效的自动化工作流。
<td>感谢 Aimzoon 对本项目的赞助！ <a href="http://aimzoon.com">Aimzoon</a> 提供稳定、高性价比的 AI API 接入服务，支持开发者将常用 AI 服务快速接入 Codex、Claude Code、Gemini CLI 等编程工具。无需复杂配置，更快接入，更稳调用，更省成本。codex倍率优惠，特价倍率等促销不断，注册即送免费体验额度，让 AI 编程真正进入日常工作流。<a href="http://aimzoon.com">点击这里</a>注册体验！
<td><a href="https://nagora.ai/">Nagora</a> 是专为开发者和团队打造的多模型 AI API 网关。通过一个账户和一枚 API Key，即可统一调用 26+ 款主流文本与图像模型，兼容 OpenAI、Anthropic 与 Gemini 协议，并可无缝接入 Claude Code、Codex、Gemini CLI 等开发工具。平台提供智能路由、自动故障转移、透明计费与统一账单，同时支持预算、限速、并发控制，让个人开发、团队协作和生产环境中的 AI 调用更稳定、更可控。无需改造现有应用，只需替换 Base URL 与 API Key，最快 1 分钟即可完成接入。</td>
<td>感谢 <a href="https://www.novada.com/?sub2api/">Novada</a> 赞助本项目！Novada 为构建 AI 应用与自动化工作流的开发者提供住宅代理、ISP 代理、数据中心代理与移动代理，以及 Web Unlocker 和 Scraper API。凭借全球 IP 覆盖、灵活的轮换与粘性会话以及精准的地理定位，Novada 帮助团队在 AI Agent 工作流、跨区域测试、网络调研与浏览器自动化等场景中稳定获取网络数据。立即体验 Novada，构建更稳定、更可扩展的 AI 工作流。</td>
<td><a href="https://anpin.ai">anpin.ai</a> のご支援に感謝します！anpin.ai は AI の普及推進に取り組むハイエンド AI 中継サービスプラットフォームです。先進的な技術アーキテクチャとグローバル分散デプロイメントにより、国際トップクラスの大規模モデルへの高速直通ルートを提供します。<br>
<td>Proxy4Free のご支援に感謝します！Proxy4Free は開発者と AI アプリケーション向けのデータプロキシサービスプロバイダーで、住宅プロキシ、静的住宅プロキシ、ISP プロキシ、データセンタープロキシなど多様なプロキシソリューションを提供しており、Web Scraping、Browser Automation、AI Agent などのシナリオに適しています。グローバル IP リソース、安定した接続、柔軟な切り替えをサポートし、開発者のデータ収集成功率の向上と IP ブロックリスクの低減を支援します。<a href="https://www.proxy4free.com/?keyword=4yjqecpc">こちらのリンクから登録</a>して、より安定した効率的な自動化ワークフローを簡単に構築しましょう。
<td>Novada のご支援に感謝します！<a href="https://www.novada.com/?sub2api/">Novada</a> は、AI アプリケーションや自動化ワークフローを構築する開発者向けに、レジデンシャル、ISP、データセンター、モバイルプロキシに加え、Web Unlocker と Scraper API を提供しています。グローバルな IP カバレッジ、柔軟なローテーション／スティッキーセッション、精密なジオターゲティングにより、AI エージェントワークフロー、クロスリージョンテスト、Web リサーチ、ブラウザ自動化などのシーンで、チームが Web データへ確実にアクセスできるよう支援します。Novada で、より安定しスケーラブルな AI ワークフローを構築しましょう。</td>
{name:"standard invalid key",item:candidate{body:`{"code":"INVALID_API_KEY","message":"Invalid API key"}`},reason:"invalid_key",match:true},
{name:"google missing key",item:candidate{body:`{"error":{"code":401,"message":"API key is required","status":"UNAUTHENTICATED"}}`},reason:"missing_key",match:true},
{name:"google group deleted",item:candidate{body:`{"error":{"code":403,"message":"API Key 所属分组已删除","status":"PERMISSION_DENIED"}}`},reason:"group_deleted",match:true},
{name:"ip acl",item:candidate{body:`{"code":"ACCESS_DENIED","message":"Access denied. Your IP is 192.0.2.1"}`},reason:"ip_acl_denied",match:true},
{name:"user not found remains",item:candidate{body:`{"code":"USER_NOT_FOUND","message":"User associated with API key not found"}`},match:false},
{name:"database failure remains",item:candidate{statusCode:500,message:"Failed to validate API key",body:`{"code":"INTERNAL_ERROR","message":"Failed to validate API key"}`},match:false},
// AudioRealtimePricePerMin applies equality check predicate on the "audio_realtime_price_per_min" field. It's identical to AudioRealtimePricePerMinEQ.
// AudioTtsPricePerMillionChars applies equality check predicate on the "audio_tts_price_per_million_chars" field. It's identical to AudioTtsPricePerMillionCharsEQ.
// LongContextPricingEnabled applies equality check predicate on the "long_context_pricing_enabled" field. It's identical to LongContextPricingEnabledEQ.
return&ValidationError{Name:"audio_realtime_price_per_min",err:fmt.Errorf(`ent: validator failed for field "Group.audio_realtime_price_per_min": %w`,err)}
return&ValidationError{Name:"audio_tts_price_per_million_chars",err:fmt.Errorf(`ent: validator failed for field "Group.audio_tts_price_per_million_chars": %w`,err)}
return&ValidationError{Name:"audio_realtime_price_per_min",err:fmt.Errorf(`ent: validator failed for field "Group.audio_realtime_price_per_min": %w`,err)}
return&ValidationError{Name:"audio_tts_price_per_million_chars",err:fmt.Errorf(`ent: validator failed for field "Group.audio_tts_price_per_million_chars": %w`,err)}
// channelmonitorrequesttemplate.BodyOverrideModeValidator is a validator for the "body_override_mode" field. It is called by the builders before save.
// group.AudioTtsPricePerMillionCharsValidator is a validator for the "audio_tts_price_per_million_chars" field. It is called by the builders before save.
Some files were not shown because too many files have changed in this diff
Show More
Reference in New Issue
Block a user
Blocking a user prevents them from interacting with repositories, such as opening or commenting on pull requests or issues. Learn more about blocking a user.
shaw
