Nightly nightly-2026-11-07

What's Changed

• port(cursor): preserve AgentService followups by @bloodf in #1659

Full Changelog: nightly-2026-11-05...nightly-2026-11-07

Nightly nightly-2026-11-06

What's Changed

• fix(release): pack the cli package, not the repository root by @bloodf in #1620
• fix(sarvam): correct the free-credit notice and declare documented context sizes by @bloodf in #1624
• fix(protocol): finish #1089 Claude restore, unanswered calls and gateway fixtures by @bloodf in #1622
• fix(media): list only dispatchable media models under routable ids by @bloodf in #1621
• feat(monitoring): add authenticated aggregate metrics export by @bloodf in #1629
• fix(providers): distinguish catalog and inference evidence by @bloodf in #1632
• fix(image): reject known non-image models by @bloodf in #1633
• fix(admission): release slots after disconnect race by @bloodf in #1634
• fix(cli): filter non-chat models from combo creation by @bloodf in #1635
• fix(qa): include shared login mark in build context by @bloodf in #1638
• fix(ui): show synced image models in provider examples by @bloodf in #1637
• fix(a11y): enlarge login interactive targets by @bloodf in #1640
• fix(stt): avoid rebilling malformed Deepgram responses by @bloodf in #1643
• ci: prepare local-first validation with manual Actions fallbacks by @bloodf in #1650
• feat(monitoring): expose aggregate admission queue depth by @bloodf in #1645

Full Changelog: nightly-2026-11-04...nightly-2026-11-06

Nightly nightly-2026-11-05

What's Changed

• fix(release): pack the cli package, not the repository root by @bloodf in #1620
• fix(sarvam): correct the free-credit notice and declare documented context sizes by @bloodf in #1624
• fix(protocol): finish #1089 Claude restore, unanswered calls and gateway fixtures by @bloodf in #1622
• fix(media): list only dispatchable media models under routable ids by @bloodf in #1621
• feat(monitoring): add authenticated aggregate metrics export by @bloodf in #1629
• fix(providers): distinguish catalog and inference evidence by @bloodf in #1632
• fix(image): reject known non-image models by @bloodf in #1633
• fix(admission): release slots after disconnect race by @bloodf in #1634
• fix(cli): filter non-chat models from combo creation by @bloodf in #1635
• fix(qa): include shared login mark in build context by @bloodf in #1638
• fix(ui): show synced image models in provider examples by @bloodf in #1637
• fix(a11y): enlarge login interactive targets by @bloodf in #1640
• fix(stt): avoid rebilling malformed Deepgram responses by @bloodf in #1643
• ci: prepare local-first validation with manual Actions fallbacks by @bloodf in #1650
• feat(monitoring): expose aggregate admission queue depth by @bloodf in #1645

Full Changelog: nightly-2026-11-04...nightly-2026-11-05

Nightly nightly-2026-11-04

What's Changed

• test(protocol): complete cross-format conformance for Gemini clients by @bloodf in #1549
• fix(thinking): per-transport thinking format for OpenCode Go /responses by @bloodf in #1576
• fix(nous): add API-key attribution tags for the Nous host by @bloodf in #1577
• fix(sarvam): drop unserved sarvam-30b and unverified limits, fetch live catalog by @bloodf in #1578
• fix(usage): record accepted image jobs and keep chat rows with odd totals by @bloodf in #1573
• test(release): smoke packed npm tarball and standalone before publish by @bloodf in #1574
• port(omniroute): add pioneer provider on the Fastino host by @bloodf in #1580
• port(omniroute): add free-ai provider by @bloodf in #1583
• chore(docs): regenerate generated files after merge batch by @bloodf in #1590
• test(security): add route authorization matrix and require one native resource id by @bloodf in #1572
• feat(routing): include nested tool-output images in capability admission by @bloodf in #1588
• fix(antigravity): floor thinkingLevel minimal to low for Gemini 3.6-3.8 Flash tiers by @bloodf in #1575
• port(omniroute): add deepai image provider by @bloodf in #1584
• fix(security): scope media-provider voice reads and record voice clone ownership by @bloodf in #1581
• feat(stt): add Codex OAuth transcription via ffmpeg by @bloodf in #1595
• fix(kiro): stop sending deprecated agent fields and update the affinity contract by @bloodf in #1591
• fix(qoder): replay queued 10605 requests instead of blocking as billing by @bloodf in #1593
• docs(deployment): add Node, driver and platform compatibility matrix by @bloodf in #1598
• feat(codex): list gpt-6.1-terra by @bloodf in #1587
• chore(docs): regenerate generated files after merge batch by @bloodf in #1607
• port(omniroute): add MLX local runtime provider by @bloodf in #1604
• test(protocol): complete cross-format gateway fixtures and Claude matrix by @bloodf in #1600
• feat(guardrails): mask PII and credentials in provider responses, including streaming by @bloodf in #1605
• fix(claude): drop blank thinking blocks and pin haiku-5-5 routing to /messages by @bloodf in #1603
• port(omniroute): add codex-app-server provider (unverified) by @bloodf in #1606
• fix(usage): record async media jobs at creation and stop combos after accepted jobs by @bloodf in #1611
• feat(stt): add Deepgram diarization and verbose_json segments by @bloodf in #1615
• fix(usage): finish paid usage accounting with slow-job health and unknown-cost counts by @bloodf in #1616
• feat(catalog): Haiku 5.5 default effort and Anthropic lifecycle on /v1/models by @bloodf in #1612

Full Changelog: nightly-2026-11-03...nightly-2026-11-04

Nightly nightly-2026-11-03

What's Changed

• chore(runtime): upgrade Node to 26 LTS by @bloodf in #1487
• fix(auth): require operator principal for key and provider-node management by @bloodf in #1417
• port(upstream): 56dbf46 - preserve non-stream SSE error results by @bloodf in #1395
• port(upstream): 9500965 - convert Claude tool-result image parts by @bloodf in #1393
• fix(translator): clamp Responses→Chat tool names to 64 chars (#1345) by @bloodf in #1384
• port(upstream): 08ef047 - bump undici to >=7.30.1 by @bloodf in #1375
• feat(openai): add gpt-6.1-terra and flag deprecated audio models by @bloodf in #1414
• fix(build): never use the filesystem root as the Next tracing root by @bloodf in #1415
• test(e2e): point API key and timeline specs at current pages by @bloodf in #1416
• port(upstream): normalize NVIDIA assistant reasoning aliases by @bloodf in #1402
• feat(claude-code): upgrade client fingerprint to 2.1.312 by @bloodf in #1372
• chore(docs): regenerate generated files after merge batch by @bloodf in #1504
• test(protocol): add Claude Messages conformance matrix for #1089 by @bloodf in #1325
• fix(iflow): reuse connection rows for the same BXAuth session by @bloodf in #1309
• fix(translator): stop openai-to-gemini mutating caller tool schemas by @bloodf in #1409
• fix(fallback): match terminal billing text only in error.message on 400 by @bloodf in #1410
• fix(translator): close Responses to Gemini usage and tool-call parity gaps by @bloodf in #1411
• feat(minimax): add GA MiniMax-M3.1-Flash and flag retiring models by @bloodf in #1413
• port(upstream): 5bdcd5e - move OpenCode free tier to /inference/v1 with the v2 fingerprint by @bloodf in #1424
• port(upstream): 95c9d03e - fill missing tool-call ids on streamed passthrough fragments by @bloodf in #1429
• port(upstream): b833128+313559f - OpenCode free reasoning for MiMo and DeepSeek flash by @bloodf in #1426
• fix(translator): restore Chat tool-name aliases on the forced-SSE JSON path by @bloodf in #1430
• port(upstream): 792f4719 a440ef43 - headroom native CommandCode, Gemini, Responses outputs by @bloodf in #1423
• port(upstream): 8a842043 - add user-authored custom system prompt by @bloodf in #1432
• fix(translator): lower Chat custom tool declarations to the Responses shape by @bloodf in #1419
• fix(translator): keep custom tool call history on Responses to Chat by @bloodf in #1420
• fix(stream): end translated nested response.failed as an error terminal and redact [Error] text by @bloodf in #1421
• fix(translator): end Gemini promptFeedback errors as error terminals and redact persisted errors by @bloodf in #1422
• chore(docs): regenerate generated files after merge batch by @bloodf in #1506
• test(protocol): tighten Responses matrix assertions (Gemini, Chat chunks, Claude delta) by @bloodf in #1438
• port(upstream): f172c3c4 - Antigravity hub/2.24.0 client fingerprint by @bloodf in #1428
• fix(cli): guard os.networkInterfaces in getLanIp without hiding network-exposed warning by @bloodf in #1440
• fix(translator): validate tool schemas and reject non-object Claude schemas by @bloodf in #1435
• fix(db): harden requestDetailsRepo handoff and dispose edge cases by @bloodf in #1439
• port(upstream): a61fc6a0 - rewrite all Hermes identity variants for Antigravity by @bloodf in #1451
• port(upstream): 2924e055 - map deepseek-v4.1-flash to upstream deepseek-flash by @bloodf in #1458
• port(upstream): f5f88039 - advertise PDF input for current Claude models by @bloodf in #1459
• port(upstream): 3ce2b828 - normalize Anthropic ids onto Kiro aliases by @bloodf in #1461
• port(upstream): 51a1827e+dbdbb648 - Copilot embeddings endpoint and catalog refresh by @bloodf in #1462
• chore(docs): regenerate generated files after merge batch by @bloodf in #1509
• port(upstream): 3fd925ab+63de82b8 - CodeBuddy CN channel headers and catalog sync by @bloodf in #1463
• port(upstream): b1effa03+a3bdaf6d - Command Code CLI 1.88.0 headers, Kimi K3.1 Pro by @bloodf in #1464
• port(upstream): b2aaf83b - surface Qoder add-on quota and never-expires plans by @bloodf in #1465
• chore(docs): regenerate generated files after merge batch by @bloodf in #1511
• port(omniroute): add bounded Responses tool-output compression by @bloodf in #1449
• port(upstream): 161e3493 - Gemini image reference input and Antigravity image error status by @bloodf in #1448
• port(upstream): 783e271c+82aeb907 - Antigravity daily generation host and effort tiers by @bloodf in #1437
• port(upstream): 29fc880e+eb373017 - Antigravity Gemini 4 Argon, Claude 5.5 rows and Argon rules by @bloodf in #1442
• port(omniroute): add side-by-side playground comparisons by @bloodf in #1481
• port(upstream): b677bfce - store model IDs in TUI combos by @bloodf in #1447
• port(omniroute): add Electron Hub, EUrouter, FastRouter, FreeInference providers by @bloodf in #1456
• port(omniroute): add AnyAPI, Apmix, Auriko, BeatAPI providers by @bloodf in #1453
• port(omniroute): add ChatAnywhere, CloudCode.ONE, Dahl, DXNT providers by @bloodf in #1455
• port(omniroute): add FreeTheAi, GreenPT, HelixMind, LiteRouter providers by @bloodf in #1457
• port(omniroute): add LLM Kiwi, LLM Gateway, Logfare, Lyceum providers by @bloodf in #1460
• port(omniroute): add Agnes, Agnes CN, AINative, Aion providers by @bloodf in #1452
• fix(providers): send registry auth and pinned headers on modelsFetcher discovery by @bloodf in #1495
• port(omniroute): add Zylo API provider by @bloodf in #1494
• chore(docs): regenerate generated files after merge batch by @bloodf in #1521
• fix(translator): parse Claude SSE on forced-stream non-streaming requests by @bloodf in #1518
• port(omniroute): add MegaNova AI, Mixlayer, Naga.ac, NaraRouter providers by @bloodf in #1488
• port(omniroute): add Openference, OpenVecta, Opper, Poixe AI providers by @bloodf in #1490
• port(omniroute): add TokenReply, UC Direct, Unifically, UnoRouter providers by @bloodf in #1492
• port(omniroute): add Cheaper Inference provider by @bloodf in #1496
• port(omniroute): add 1min.ai provider by @bloodf in #1497
• port(omniroute): add Magnific Mystic image provider by @bloodf in #1499
• port(omniroute): add signed operator event webhooks by @bloodf in #1482
• chore(docs): regenerate generated files after merge batch by @bloodf in #1522
• port(omniroute): add single-instance Kubernetes manifests by @bloodf in #1485
• fix(translator): validate and coerce Responses tool declaration schemas by @bloodf in #1486
• port(upstream): 5cdcddf5 - advance combo past a request-scoped 4xx by @bloodf in #1431
• port(upstream): 7c42d8f - normalize OpenAI completion token caps by @bloodf in #1394
• port(upstream): ce060c79 - OpenCode free-tier TLS client via optional wreq-js by @bloodf in #1483
• test(protocol): assert #1330 non-stream error finish as shared 502 contract by @bloodf in #1373
• test(chat): cover account cooldown and fallback for SSE error envelopes in non-stream requests by @bloodf in #1445
• chore: update test package vitest and happy-dom to clear audit...

Nightly nightly-2026-11-02

What's Changed

• ci: report test progress and clean up timed-out worker fixtures by @bloodf in #1400
• fix(startup): tolerate unavailable network interface enumeration by @bloodf in #1397

Full Changelog: nightly-2026-10-31...nightly-2026-11-02

Nightly nightly-2026-11-01

What's Changed

• fix(website): improve marketing and documentation workflows by @bloodf in #1392

Full Changelog: nightly-2026-10-30...nightly-2026-11-01

Nightly nightly-2026-10-31

What's Changed

• fix(website): improve marketing and documentation workflows by @bloodf in #1392

Full Changelog: nightly-2026-10-30...nightly-2026-10-31

Nightly nightly-2026-10-30

What's Changed

• fix(translator): preserve all Gemini system messages by @bloodf in #1261
• ci(fleet): add scoped local gates with full checkpoints by @bloodf in #1279
• ci(fleet): scope upstream watch helper checks by @bloodf in #1282
• port(upstream): #4532 - prevent generated Gemini tool ID collisions by @bloodf in #1281
• fix(upstream): qualify ported ledger membership by source by @bloodf in #1278
• fix(executors): prevent unknown providers from falling through to OpenAI by @bloodf in #1283
• fix(mcp): integrate diagnostic privacy boundaries and regressions by @bloodf in #1284
• fix(provider): separate Ollama reachability from credential verification by @bloodf in #1285
• fix(cursor): reject unenforceable output caps by @bloodf in #1264
• ci: supervise full test inventory in sequential bounded batches by @bloodf in #1299
• fix(translator): reject unsupported audio blocks by @bloodf in #1259
• fix(translator): preserve client system authority by @bloodf in #1257
• fix(translator): preserve tool failure semantics by @bloodf in #1260
• fix(translator): protect opaque thinking continuity by @bloodf in #1258
• fix(crypto): publish first-start keys atomically by @bloodf in #1265
• fix(privacy): redact MiMo login diagnostics by @bloodf in #1267
• fix(audit): derive provider identities from exports by @bloodf in #1266
• fix(cursor): reject unsupported image input explicitly by @bloodf in #1262
• fix(kiro): reject unverified output-limit enforcement by @bloodf in #1263
• fix(kiro): reject unresolved remote image loss by @bloodf in #1255
• chore: configure Renovate with preserved onboarding history by @bloodf in #1301
• ci(nightly): install website dependencies before the Vitest gate by @bloodf in #1307
• fix(translator): integrate Responses media validation and tool-output compatibility by @bloodf in #1297
• fix(providers): integrate discovery and provider boundary regressions by @bloodf in #1298
• port(upstream): 625df74 - sanitize native Antigravity tool responses by @bloodf in #1280
• fix(security): enforce native resource ownership and encoded routing by @bloodf in #1271
• chore(deps): update dependency undici to v7.30.1 [security] by @renovate[bot] in #1316
• chore(deps): update dependency storybook to v10.6.2 [security] by @renovate[bot] in #1315
• fix(deps): update vulnerable root and website dependency closures by @bloodf in #1296
• fix(mcp): sanitize rendered retry and list diagnostics by @bloodf in #1270
• ci: confine local-CI child temporary files to a private root by @bloodf in #1306
• fix(auth): preserve reauth state during in-flight cooldown updates by @bloodf in #1308
• chore(security): fix #1271 follow-up nits in owner JSDoc and test helper by @bloodf in #1321
• fix(auth): keep reauth when the atomic fallback write throws by @bloodf in #1326
• fix(stream): redact translated upstream errors and preserve passthrough errors by @bloodf in #1276
• test(protocol): expand Chat Completions conformance matrix for #1089 by @bloodf in #1322
• fix(translator): map Chat tool_choice strings to Gemini functionCallingConfig modes by @bloodf in #1337
• test(protocol): finish Responses API conformance matrix for #1089 by @bloodf in #1323
• fix(chatCore): return 502 for SSE→JSON error finish like non-stream by @bloodf in #1348
• fix(protocol): normalize empty Claude tool input_schema to an object schema by @bloodf in #1351
• docs(upstream): record 9router review through 1fe586bb (v0.5.112) by @bloodf in #1370
• fix(translator): preserve Chat tool_choice when lowering to Responses by @bloodf in #1336
• fix(protocol): normalize Claude tool schemas on every Claude-bound path by @bloodf in #1379
• port(upstream): 8db1f4a - enforce Gemini protobuf type predicates on schemas by @bloodf in #1371
• fix(translator): end Responses streams with incomplete/failed for length/error finishes by @bloodf in #1347
• port(upstream): 5a4e812 - strip NUL bytes from bound DB strings by @bloodf in #1376
• feat(providers): add Kimi Platform K3.1 Pro, Highspeed, and K2 catalog models by @bloodf in #1377
• fix(translator): pair tool results by call id within the current assistant turn by @bloodf in #1380
• fix(translator): project non-stream Responses bodies to Gemini shape by @bloodf in #1385
• port(upstream): 4662724 - count Ollama prompt_eval_cached_count as cached tokens by @bloodf in #1386
• port(upstream): b82f5c4 - treat credit exhaustion as backoff so combos try next model by @bloodf in #1388

New Contributors

• @renovate[bot] made their first contribution in #1316

Full Changelog: v4.11.0...nightly-2026-10-30

v4.11.0

4.11.0

Features

• feat: integrate DurinDoor 4.11 improvements
• feat(website): expand the hero scene and publish the brand guide
• feat(gateway): add MCP presets and PostgreSQL-only operation (#1049)
• feat(web): rebuild DurinDoor identity, website, and documentation

Fixes

• fix(website): restore the dimensional gateway animation

Maintenance

• docs: separate user guides from comprehensive AI references (#1053)