Skip to content

Merge upstream QuantumNous/new-api (50 commits) into main - #8

Merged
mrdjango merged 94 commits into
mainfrom
merge/upstream-20260917
Sep 17, 2026
Merged

mrdjango merged 94 commits into
mainfrom
merge/upstream-20260917

Conversation

@mrdjango

@mrdjango mrdjango commented Sep 17, 2026 •

Copy link
Copy Markdown
Owner

PR Notice

Important

Merge this PR with "Create a merge commit", not "Squash and merge".
PR #6 was squash-merged, which dropped upstream as a parent. main therefore still looked 92 commits behind upstream, 42 of them already absorbed. The first commit here (a5cd6a030) restores that ancestry without changing any file. Squashing this PR would discard it again, and the next sync would hit the same problem.

This PR was prepared with AI assistance (Claude Code). The author is not one of the upstream core developers.

Related Issue

  • N/A — periodic sync of the TensorGrid fork with upstream QuantumNous/new-api.

Type of change

  • Bug fix
  • New feature
  • Performance / Refactor
  • Documentation

The new features come from upstream; the bug fixes are billing issues this merge exposed (below).

Description

Merges upstream ea7cb0ba4..69a500298 (50 commits) into main and reconciles it with the TensorGrid changes.

Conflicts

File Resolution
common/model.go Upstream independently shipped our gpt-image- fix plus more image models. Took upstream.
common/model_test.go (add/add) Kept upstream's common_test package and moved our IsImageGenerationModel table into it.
model/main.go Kept both new migration steps: upstream's options primary-key repair, then our outbox width fix. Dropped migrateDBFast: upstream deleted it in 2b6f1dfef, an earlier fork merge kept an edited copy, and nothing calls it.
service/billing_session.go TensorGrid absolute-target reservation first, then upstream's new image reservation path.
controller/relay_task_plugin_test.go Kept both sides' tests.
.gitignore Kept graphify-out/.

Billing fixes the merge needed

  1. Image refund lost the earlier pre-consume (upstream bug, non-TensorGrid wallets). Upstream's new image path (f064bffa2) calls WalletFunding.PreConsume(delta) to top up a reservation. PreConsume assigned consumed, and Refund returns exactly consumed. So a failed image request refunded only the top-up. Reproduced on pristine upstream: a 50k pre-consume plus a 50k top-up refunded 50k, and the user lost the other 50k. PreConsume now accumulates. TensorGrid wallets go through their own absolute-target branch and were not affected.
  2. A refunded session could be settled again. Our fork refunds tasks that fail at submission instead of settling them at zero. Upstream's new TestImmediateTaskSettlementDatabase then settles the same session, and settle credited the pre-consume a second time. settle is now a no-op once the session has been refunded, mirroring how Refund already ignores a settled session. No production path settles a refunded session today, so this is defense in depth.
  3. RecordConsumeLog read the user cache when logging was off. It still enqueues the TensorGrid credit event when consume logging is disabled, but it now reads the IP-logging setting only when it actually writes the user log. The old read panicked a test that disables logging while Redis is unset; the same latent failure already existed on main.
  4. Immediate task failure keeps TensorGrid semantics. Upstream now settles failed-at-submit tasks at zero and writes a zero-quota consume log. Our fork keeps the refund path the control plane already handles. In TestImmediateTaskSettlementDatabase, only the failed case changed: it asserts the balance is restored once the asynchronous refund lands, and that no consume log is written. The other seven cases are unchanged.

Regression tests: TestBillingSessionImageReserveRefundReturnsWholeReservation and TestBillingSessionSettleAfterRefundMovesNoQuota in service/tiered_settle_test.go.

TensorGrid integration audit

Checked by hand rather than trusting the auto-merge:

  • The credit-event enqueue, and consumeParams threading on every SettleBilling / PostConsumeQuota / RecordConsumeLog call site.
  • Wallet reserve/refund/adjust hooks, IsTensorGridUser gating, and the immediate-failure refund path.
  • The internal-route rate-limit exemption and 429 JSON envelope from Exempt the TensorGrid internal API from the global IP rate limit #7.
  • The audio-transcription endpoint type (including upstream's reworked advanced custom routes) and the four fork usage keys.

Upstream's new files relay/request_billing.go and service/image_billing.go route through the hooked PreConsumeBilling / Reserve / Refund path. The new Responses WebSocket relay (9fe0457ee) bills each response.create turn like an HTTP streaming Responses call and settles through PostTextConsumeQuota. So it inherits the TensorGrid hooks and doesn't need the realtime socket's forced pre-consume.

Proof of Work

Check Result
go build ./... / go vet ./... pass
go test ./... 42 packages, 0 failures
cd relaykit && GOWORK=off go vet/build/test ./... pass
CI gofmt -l on TensorGrid files, and on every file I edited clean
bun run typecheck pass
bun run test 136 files, 1,697 tests, all pass
bun run build pass

Runtime. I built both binaries with the image's flags (CGO_ENABLED=0 GOWORK=off, linux/amd64) and booted them with the required TensorGrid environment:

  • Fresh database, two boots: /api/status 200, frontend 200, /api/internal/tensorgrid/v1/catalog 401 without a signature, no FATAL, panic or [ERR] lines.
  • Upgrade: pre-merge main created the database. I seeded a TensorGrid user (wallet 3,000,000,000, above the int32 range), an account, an outbox row with a 179-character event_id, and an options row. The merged binary then booted twice. All seeded rows survived both boots unchanged. The upgraded schema matches a fresh merged schema exactly: 43 tables, 164 indexes, no column differences.

Database verification: SQLite only. The app uses its embedded glebarez/sqlite; the test suite reports SQLite 3.50.4. MySQL and PostgreSQL were not verified: Docker was not running and neither server was installed. The schema changes in this merge are upstream's (upstream's commit messages report SQLite, MySQL and PostgreSQL runs), plus our unchanged outbox migration. Per AGENTS.md this PR does not claim MySQL/PostgreSQL compatibility; please restart once against a real MySQL or PostgreSQL staging database before or after merging.

Known upstream issues, unchanged: TestSecurityAccountDeletionConcurrentRequestsHaveOneWinner still flakes about 1 run in 3 (it passed this run), and web/src/features/pricing/constants.ts:19 still has upstream's no-import-type-side-effects lint error.

Checklist

  • Human review: Whether or not the description was AI-generated, I have reviewed the full content and take responsibility for its accuracy and completeness.
  • Not a duplicate: N/A — fork sync.
  • Feature issue: N/A — the features are upstream's.
  • Prior discussion: N/A — routine fork sync.
  • Scope: This PR is not a Coding Plan, reverse-engineered channel, third-party API wrapper, or a change to the Codex channel type.
  • Focused change: An upstream sync plus the billing fixes it required; not a single focused change.
  • Local verification: I have run tests or manually verified locally so maintainers can re-check from this evidence.
  • Security: This change does not include secrets and follows the project's coding guidelines.

🤖 Generated with Claude Code

Calcium-Ion and others added 30 commits September 4, 2026 10:07
Model-name post-processing is rebuilt around an explicit trailing
@key:value modifier syntax (thinking/effort/temperature/topp) that
overrides request fields, survives model mapping, and records
conversion diagnostics on the consume log.

- Legacy naked aliases (-thinking, -nothinking, -thinking-<budget>,
  effort tails) now parse only for positively matched families
  (gpt-*/o-series, claude-*, gemini-*, incl. vendor/ namespaces);
  names like qwen-max stay opaque. EffortTailModelIDs remains the
  escape hatch for real in-family IDs such as gpt-5.1-codex-max.
- Billing identity resolves once in ModelPriceHelper via a ladder:
  configured request name first (legacy wildcard entries intact), then
  canonical billing names rebuilt from parsed intent
  (base@effort:E@thinking:S, then base@thinking:S; order, duplicates,
  and budget values are irrelevant; temperature/topp never priced),
  then base. Routing and token limits fall back through
  RoutingMatchModelName; pricing lookups stay wildcard-only.
- Pass-through stays byte-identical: modifiers and aliases are neither
  parsed nor validated there and forward verbatim for the upstream
  (or a chained gateway) to interpret.
- Unknown modifier keys and invalid known-key values are rejected with
  400; models whose real names contain @tag:value are exempted via the
  thinking-suffix blacklist, which now supports re:-prefixed Go regex
  entries.
- Claude reasoning render coerces unsupported combinations (disable,
  adaptive, budgets) with warning diagnostics instead of erroring;
  native-protocol requests without host syntax pass through untouched.

BREAKING(openrouter): drop the host-invented "-thinking" model-name
alias (added in 4f6d16e) that trimmed any *-thinking model on
OpenRouter channels and injected reasoning.enabled. It matched too
broadly and mangled real model IDs such as kimi-k2-thinking.
Migration: use some-model@thinking:on, or keep the old public name via
a channel model mapping {"some-model-thinking": "some-model@thinking:on"}.
Claude/Gemini family aliases (incl. anthropic/claude-*-thinking) keep
working via the family whitelist.
* perf(relay): bulk-copy Responses raw JSON fields

* perf(common): share RawMessage deep-copy optimization

---------

Co-authored-by: CaIon <i@caion.me>
…s#7211)

* fix(relay): treat gpt-5 and later generations alike for max_completion_tokens

IsOpenAIGPT5Model matched on the literal prefix "gpt-5", so gpt-6-astra
(and every generation after it) fell through the gpt-5 request rules:
max_tokens was forwarded as-is and the provider rejected it with
"Unsupported parameter: 'max_tokens' is not supported with this model.
Use 'max_completion_tokens' instead." The same gap left temperature,
top_p and logprobs untouched, each of which the provider also rejects,
and made the channel test button report a 400 for a healthy deployment.

Match on the major version instead (gpt-<n>... with n >= 5). Callers are
unchanged: ConvertOpenAIRequest, GetSystemRoleName, buildTestRequest and
the health check all go through this one helper. buildTestRequest now
sends max_completion_tokens for these models directly instead of relying
on the later conversion. gpt-4.1, gpt-4o, gpt-oss, gpt-image and
gpt-realtime names still do not match.

Verified against Azure OpenAI gpt-6-astra (2026-09-03): with the old
prefix max_tokens / temperature / top_p / logprobs each returned 400,
while gpt-5.6-luna with the same payload returned 200.

Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_016bwz8o5UeoRtrtDusKaayp

* fix(relay): separate OpenAI chat model compatibility rules

---------

Co-authored-by: Claude Fable 5.1 <noreply@anthropic.com>
Co-authored-by: CaIon <i@caion.me>
Move account security settings into a dedicated page and add token status, rotation, revocation, and access history.

Store audit events with role snapshots and JSON metadata, add audit.read authorization and an independent audit page, and upgrade the ClickHouse driver to v2.46.0.
Upgrade the SQLite GORM driver to correctly distinguish unique indexes from constraints. Normalize equivalent MySQL decimal defaults and PostgreSQL CHAR metadata during migration comparison while preserving real schema changes.

Validation: 27 fresh-start and rc.26 upgrade scenarios using SQLite 3.50.4, MySQL 5.7.44, PostgreSQL 9.6.24, and ClickHouse 25.8.33.6; 15 upgraded databases passed uniqueness checks. Relational databases issue no DDL on unchanged restarts. Existing ClickHouse TTL synchronization remains unchanged.
Require single-use operation proofs for passkey enrollment, two-factor setup, and channel key access. Add password and OAuth verification flows, enforce session-bound enrollment, and redact OAuth callback secrets from logs.

Validation: affected Go packages pass; frontend typecheck, changed-file lint, and 111 tests pass. Security enrollment regressions pass on SQLite 3.50.4, MySQL 8.0.46, and PostgreSQL 16.15. Full frontend lint has pre-existing errors outside the changed files.
Use authorization code flow with PKCE and verified ID tokens for Telegram login, binding, and security verification. Preserve existing bindings and require administrator OAuth configuration.

Keep the restricted WeChat first-enrollment session proof, fix missing-target authentication errors, and preserve callback requests after OAuth popups close.
Require scoped, single-use verification for account bindings and password
operations. Bind OAuth authorization and email confirmations to the initiating
session; preserve the last usable login method and audit operation outcomes.

Apply Unicode-aware password length limits, Argon2id writes with bcrypt
compatibility, and long-password encryption.
Return has_password with the existing profile SELECT without extra queries.
Reuse the existing security dialogs and add all seven locale translations.

Validation:
- Go: go test ./common ./model ./service ./middleware ./controller ./router -count=1
- DB: SQLite 3.50.4, MySQL 8.4.11, PostgreSQL 16.15; separate main/log databases
- MySQL/PostgreSQL: TEST_SECURITY_DIALECT=<dialect> with TEST_<DIALECT>_DSN,
  go test ./controller -run '^(TestSecurityAccount|TestSecurityEnrollment|TestGenerateOAuthCode|TestOAuthBind|TestTelegramOAuth)' -count=1 -v
- Web: relevant Vitest suites, bun run typecheck, targeted oxlint/format,
  bun run i18n:sync, and bun run build

Roll out dual-format readers to every instance with
ACCOUNT_PASSWORD_HASH_ALGORITHM=bcrypt before enabling Argon2id writes
and the new UI. Rollbacks must retain Argon2id and v2 envelope readers.

Relevant controls: ASVS 5.0.0 6.2.1-6.2.3, 6.2.5-6.2.9, 6.3.7, 7.4.3, 7.5.1;
this change does not assert application-wide ASVS certification.
Record successful and failed API token operations with safe target metadata. Capture quota adjustments in a transaction, synchronize committed cache differences, and correlate audit and top-up records.

Show operation targets, changes, quota balances, and failure details consistently across audit and usage logs, with translations for all seven locales.

Validated controller, middleware, and model tests; 78 frontend tests; typecheck and lint; real SQLite 3.50.4, MySQL 8.4.11, and PostgreSQL 16.15 with shared and separate log databases.
Treat TOTP and Passkey as alternative enrolled factors across login and
sensitive account operations. Gate every primary login transport before
issuing a session, require WebAuthn user verification, and consume login
challenges atomically with session creation.

Reuse the shared verification UI for login, 2FA management, and account
deletion. Require scoped, single-use deletion proof; recheck the session
inside the deletion transaction and revoke all sessions afterward.

Validation: controller/service/model/middleware tests; real SQLite 3.50.4,
MySQL 8.4.11, and PostgreSQL 16.15 security regressions; frontend tests,
TypeScript, targeted lint, formatting, and production build.

Deploy the frontend and all backend nodes together. No schema changes.
Introduce a unified model management experience: catalog metadata
validation, vendor management, batch delete with channel/pricing
cleanup, model pricing snapshot editing with optimistic concurrency,
and an upstream ratio-sync flow with price cells. Move configuration
into dedicated pricing config/metadata-sync/vendor-management backend
services and add audit records for model/vendor/pricing mutations.

Rework the models page around vendors and model connections, add
model-pricing and vendor-management dialogs, and replace the shared
Select usages with the Combobox component across subscriptions,
plugins, OAuth presets, audit filters, and settings. Add the model
pricing panel and verify behavior with focused tests.
* feat(ali): support wan3.0 all-in-one video models

Extend the Alibaba task plugin for wan3.0-video / wan3.0-video-prime
using DashScope media, resolution/ratio, duration rules, and billing ratios.

* fix(ali): make wan3.0 smart duration billable and alias-safe

Review fixes on top of QuantumNous#7240 (qiuliw):

- duration -1 never reached the plugin: the host rejects negative
  canonical duration/seconds facts before any hook runs, on every
  entry point. Decoders now normalize -1 into an auto_duration marker;
  convert emits -1 upstream and bills 30s up front; non-wan3.0 models
  reject the marker instead of silently defaulting to 5s.
- extractUsageOnComplete read output.duration/output.resolution, but
  wan3.0 reports usage.output_video_duration and numeric usage.SR, so
  smart-duration and resolution settlement never reconciled. Read the
  usage block first, keep the legacy output fields as fallback.
- convert keyed default resolution on the client model name, so a
  channel-mapped alias fell to 720P while the direct request got 1080P.
  Every model-shaped decision in convert now uses ctx.upstreamModel.
- Unknown wan3.0 size values were silently coerced to 1080P; reject them.
- Image-only openai_responses input was accepted for every model,
  regressing t2v models into pre-consume then upstream rejection.
  Restore the guard and allow image-only for i2v and wan3.0 only.
- Native passthrough dropped wan3.0 parameters (ratio, audio) unless
  media was present; forward them via metadata for wan3.0.
- Hoist the triplicated size-to-resolution map; add wan3.0 contract tests.

---------

Co-authored-by: qiuliw <a1807191473@qgmail.com>
Kimi K3 injects tools mid-conversation via a system message that carries
a `tools` array. `dto.Message` had no such field, so the tools were
silently dropped during the parse/re-marshal round trip and the upstream
rejected the request with `'tool_choice'='required' requires a 'tools'
field`.

- add `Message.Tools` (json.RawMessage passthrough)
- omit the `content` key only for tool-loading messages with nil content,
  as Kimi rejects `tools` next to `content`; all other messages keep
  emitting `"content": null`
- count message-level tools in token estimation
- skip tool-loading messages in channel system prompt injection and make
  the compatible handler reuse applySystemPromptIfNeeded
- add kimi-k3 to the moonshot model list

Fixes QuantumNous#7235
`common/json.go` and `relaykit/relayconvert/kitutil/json.go` were two
hard-wired copies of the same encoding/json wrapper, so swapping the JSON
engine required editing both modules.

- kitutil defines a `Codec` interface with a standard-library default and
  a `SetCodec` hook, mirroring the existing SetLogging host hook; every
  kitutil JSON helper and relaykit DTO (un)marshal method goes through it
- `common/json.go` forwards to kitutil and injects `hostJSONCodec` from
  init() so tests run on the same engine as production; swapping the
  engine now touches only this type in the root module
- route the remaining direct encoding/json calls inside relaykit
  (dto/values.go, responses stream validation) through kitutil
- add a codec routing test and a host codec conformance test locking the
  encoding semantics the DTOs depend on

Direct encoding/json call sites in the root module are left for a
separate cleanup.
- Portal Combobox and Select popups into the vaul DrawerContent via a
  portal-container context so they stay inside the Radix modal layer
  instead of inheriting body pointer-events: none
- Provide an in-memory localStorage/sessionStorage in test-setup when the
  Node 25+ global accessor resolves to undefined and shadows jsdom
- Add regression tests for popups rendered inside and outside the drawer
The perf summary API only returned the last three non-empty buckets as
bare success rates, so the 24-slot status strip on model cards never lit
more than three bars and could not show hours without traffic.

Replace recent_success_rates with recent_success_series: one timestamped
point per hour that had requests, aligned to hour start regardless of
the configured bucket size. The badge now anchors its 24 slots to the
client's current hour and places each point by timestamp, leaving hours
without data gray.
… plugin

Disabling a task plugin that has both a factory built-in and an override
row only flipped the override flag, so the built-in kept routing the same
models and same-name uploads (e.g. minimax-h3 vs MiniMax-H3) still hit a
routing conflict. Now the key also enters the disabled-factory set and the
list reports "disabled" instead of "disabled_fallback" when nothing serves.
…t-ins

Decode on ctx.upstreamModel || ctx.model and echo ctx.model; fix the
lyrics/music render branch that read a nonexistent ctx.requestBody.model;
stop sending empty Accept/Content-Type. Bump sunoapi to 1.0.2. Add an
alias-echo table test covering every built-in.
TaskPluginOverrideEnabled had no UI since the master switch landed, yet
when left off it marked every third-party plugin "disabled; platform
unavailable" regardless of its own toggle. Drop the option, env var,
registry flag, and the dead branch in ListTaskPlugins; the master switch
and per-plugin toggles are the only two levels now.
The model card grid and its loading skeleton only reached three columns at
2xl, leaving a two-column layout on common desktop widths. Use the xl
breakpoint for both and lock it with a test.
Calcium-Ion and others added 29 commits September 11, 2026 21:44
Preserve legacy credentials across primary RP ID changes and bind successful
assertions to their verified RP ID. Share domain selection across sign-in,
login verification and sensitive operations without replaying ceremonies.

Save domain settings atomically, preview known and unknown credential impact,
and require fresh confirmation before removing domains still in use. Recheck
trusted RP IDs when persisting credentials and record safe operation audits.

Use read-only configuration snapshots, accept trusted HTTPS intranet hosts,
check registration origins early, and preserve historical capitalization.
Update settings guidance and all supported translations.

Validation: SQLite 3.50.4, MySQL 8.0.46 and PostgreSQL 16.15, including fresh
and released-schema upgrade checks; Go regression and race checks; 79 frontend
tests; TypeScript, lint, frontend build and root Go build.

Security references: OWASP ASVS 5.0.0 V6.3.4 and V6.3.8, Authentication and
Session Management Cheat Sheets, and the WebAuthn RP ID/origin rules.
Identify legacy global uniqueness by definition instead of object name. Replace colliding target names inside the locked transaction, preserving unrelated indexes and rolling back failed migrations.

Verified fresh schemas, repeated migration, rollback, and upgrades from v1.0.0-rc.37 on SQLite 3.50.4, MySQL 5.7.44/8.0.46, and PostgreSQL 9.6.24/16.15.
Extract the usage-log filter panel into a shared data-table component built on the existing Collapsible primitive. Add an opt-in toolbar mode for channels that collapses filters while keeping quick actions accessible.

Validated 12 tests across mobile toolbar, usage-log filters and mobile layout; frontend typecheck and lint passed for all changed files.
Resolve obsolete PostgreSQL single-column unique constraints by their actual
catalog names inside a locked transaction. Preserve unrelated indexes and
roll back constraint removal when foreign keys depend on them.

Load legacy authorization policies as structured fields. Preserve scoped
rows during baseline reseeding and load unsupported own/other scopes as
deny, preventing fallback to broader admin grants. Existing global grants
remain usable; unsupported scoped grants require administrator review.

Validation: fresh databases and rc.37 through rc.27, 60/60 combinations on
SQLite 3.50.4, MySQL 5.7.44/8.0.46, and PostgreSQL 9.6.24/16.15. Each ran
InitDB, separate InitLogDB, and authz.Init twice with persisted data.
Focused schema, foreign-key rollback, and master/slave policy reload
regressions passed, along with the authz package tests.

Security references: OWASP ASVS 5.0.0 controls 8.2.1, 8.2.2, and 8.3.1;
OWASP Authorization Cheat Sheet (least privilege and deny by default).
Pass legacy RP IDs through the authentication section registry and require the defaults property to prevent omissions. Add settings-page regression coverage for empty, single, and multiple compatible domains.

Validation: 19 passkey settings tests passed; TypeScript, targeted lint, and formatting checks passed.
…ity races (QuantumNous#7367)

* test(web): stabilize flaky vitest timeouts and motion visibility races

* test(web): exclude no-preference from reduced-motion media query mock
…ntumNous#7365)

* fix(web): prevent combobox dropdowns opening on dialog autofocus

* test(web): open version selector before typing in race regression
…mNous#7321)

Uppercase or mixed-case extensions (e.g. .MP3, .WAV) fell through to the
default branch and returned "unsupported audio format", so audio token
usage could not be measured. Lowercase the extension before the switch.

Closes QuantumNous#7319
* Add Responses WebSocket relay support

* Fix Responses WebSocket relay protocol issues

* Add status to Responses WebSocket errors

* Fix Responses WebSocket local error handling

* Use Responses error payload for control write failures

* Fix WebSocket subprotocol auth handling

* fix: address responses websocket CR feedback

* fix: truncate responses websocket relay errors

* fix: preserve websocket error message

* test: cover responses usage output fallback

* feat: close active websockets for disabled channels

* chore: address websocket manager review feedback

* fix: adapt responses websocket relay after rebase

* feat: add per-channel Responses WebSocket switch

* fix: address responses websocket relay review findings

* fix: share Responses WebSocket billing with HTTP and retain existing rate limits
* feat: vllm channel

* fix: vllm status i18n

* feat: sglang channel

* fix: sglang icon

* fix: prevent long SGLang metric labels from overlapping values

* fix: coerce SGLang responses created_at to integer

* fix: ExceedsMaxTokensLimit
PR #6 was squash-merged, which dropped the upstream parent. main already
contains ea7cb0b's content (git diff ea7cb0b main lists only fork
files), so this merge changes no files; it restores the ancestry so the
next upstream merge uses ea7cb0b as its base instead of replaying the
42 commits PR #6 already absorbed.

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Brings in 50 upstream commits (ea7cb0b..69a5002) and reconciles them
with the TensorGrid fork. The preceding commit restores ea7cb0b as an
ancestor, since PR #6 was squash-merged and git would otherwise have
replayed the 42 commits it already absorbed.

Conflicts resolved:

- common/model.go: upstream independently shipped the fork's gpt-image-
  prefix fix plus more image models; took upstream.
- common/model_test.go (add/add): kept upstream's external test package
  and moved the fork's IsImageGenerationModel table into it.
- model/main.go: kept both new migration prologue steps (upstream's
  options primary-key repair, then the TensorGrid outbox width fix).
  Also dropped migrateDBFast: upstream deleted it in 2b6f1df, an
  earlier fork merge kept a TensorGrid-edited copy, and nothing calls it.
- service/billing_session.go: kept the TensorGrid absolute-target
  reservation first, then upstream's new image reservation path.
- controller/relay_task_plugin_test.go: kept both sides' tests.
- .gitignore: kept graphify-out/.

Fixes the merge needed:

- WalletFunding.PreConsume now accumulates consumed. Upstream's new
  image path calls it again with the increment, and Refund returns
  exactly consumed, so an image request that failed after topping up a
  reservation refunded only the top-up (reproduced on pristine upstream:
  a 50k + 50k reservation refunded 50k). TensorGrid wallets use their own
  absolute-target branch and were not affected.
- BillingSession.settle is a no-op once the session was refunded. The
  fork refunds tasks that fail at submission instead of settling them at
  zero; upstream's new database test then settles the same session,
  which credited the pre-consume a second time.
- RecordConsumeLog reads the user's IP-logging setting only when it
  writes the user log. It still enqueues the TensorGrid credit event
  when consume logging is off, but no longer hits the user cache for it;
  that read panicked a test that disables logging with Redis unset.
- TestImmediateTaskSettlementDatabase: the "failed" case now asserts the
  fork's refund semantics (balance restored once the asynchronous refund
  lands, no consume log) instead of upstream's zero-quota settlement.
  Every other case is unchanged.

TensorGrid integration audited after the merge: credit-event enqueue,
consumeParams threading on every settle/consume call site, wallet
reserve/refund/adjust hooks, the immediate-failure refund path, the
internal-route rate-limit exemption, the audio-transcription endpoint
type, and the four fork usage keys are all intact. Upstream's new
Responses WebSocket relay bills per turn through PrepareRequestBilling,
RefundFailedRequestBilling and PostTextConsumeQuota, so it inherits the
TensorGrid hooks without the realtime socket's forced pre-consume.

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
@mrdjango
mrdjango merged commit 34a014b into main Sep 17, 2026
2 checks passed
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.