- Fix usageExtractor cached_tokens fallback for Responses API (use cache_read_input_tokens when input_tokens_details is absent)
- Fix truncated claude-native-passthrough-tools.test.mjs that caused parse error
The second translateRequest call in the claude->openai->claude passthrough
path had an extra `translatedBody` argument before `stream`, shifting all
parameters by one. This caused the `stream` field in the upstream request
to be set to an object instead of a boolean, producing:
"stream: Input should be a valid boolean"
Co-Authored-By: Craft Agent <agents-noreply@craft.do>
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
The openai-to-claude translator was prefixing tool names with 'proxy_'
(e.g. Bash → proxy_Bash) even when routing Claude-format requests to
native Claude/Anthropic providers. Claude rejects unknown tool names,
causing 'No such tool available: proxy_Bash' errors.
Root cause: the _disableToolPrefix condition only disabled the prefix
for non-Claude providers, but it should be disabled for ALL providers
in the Claude passthrough path since tools are already in Claude format.
Fixes#618
- Rename in.json → hi.json: 'in' is Indonesian (ISO 639-1), Hindi is 'hi'.
Fixes Weblate locale conflict where id.json and in.json both claimed Indonesian.
- Move empty tool name filter before Codex passthrough: nativeCodexPassthrough
skipped all input sanitization, causing 400 'empty tool name' from upstream.
- Collapse 3+ consecutive newlines to \n\n in response sanitizer: thinking
models accumulate excessive line breaks between tool call blocks.
- Handle reasoning_details[] array (StepFun/OpenRouter format) in sanitizer and translator
- Handle 'reasoning' field alias → reasoning_content in streaming and non-streaming paths
- Cross-map input_tokens/output_tokens ↔ prompt_tokens/completion_tokens in filterUsageForFormat
- Fix extractUsage to accept input_tokens/output_tokens as alternative field names
- All 936 tests pass
## Proxy UI Bug Fixes
- fix: proxy badge on connection cards now uses resolveProxyForConnection()
per-connection (covers registry + config-file assignments)
- fix: Test Connection button now works in 'saved' proxy mode by resolving
proxy config from savedProxies list
- fix: ProxyConfigModal now calls onClose() after save/clear (fixes UI freeze)
- fix: ProxyRegistryManager loads usage eagerly on mount with deduplication
by scope+scopeId to prevent double-counting; adds per-row Test button
## Connection Tag Grouping (new feature)
- feat: add Tag/Group field to EditConnectionModal (stored in
providerSpecificData.tag, no DB schema change)
- feat: connections list groups by tag with visual dividers when any account
has a tag; untagged accounts appear first without header
## Post-merge fix from PR #607 review
- fix: function_call blocks in translateNonStreamingResponse now also strip
Claude OAuth proxy_ prefix via toolNameMap (kilo-code-bot #607 warning)
Affects OpenAI Responses API format path — tool_use was fixed in PR #607
but function_call was missed
- fix(translator): pass toolNameMap to translateNonStreamingResponse so Claude
OAuth proxy_ prefix is correctly stripped from tool_use block names in
non-streaming responses (was only stripped in streaming path)
- fix(validation): add LongCat specialty validator that probes /chat/completions
directly, bypassing the /v1/models endpoint that LongCat does not expose (#592)
Co-authored-by: diegosouzapw <diegosouzapw@users.noreply.github.com>
- Reword comments that contained the token any; replace any types with typed shapes
- stream.ts: passthrough tool-call flag via local boolean (state is null in passthrough)
- Document T11 in zws_docs/ZWS_README_V8.md
Made-with: Cursor
Cherry-pick from codex/omniroute-fixes-20260324:
- Replace MCP singleton transport with per-session architecture for Streamable HTTP
- Fix Claude passthrough via OpenAI round-trip normalization
- Add detectFormatFromEndpoint() for endpoint-aware format detection
- Support raw code#state in OAuth modal for Claude Code remote auth
- Expose cloudConfigured/cloudUrl/machineId in settings API
- Switch docker-compose.prod.yml target to runner-cli
- Add 3 new tests for round-trip and detectFormat
PR: #562
fix(chat): convert tool_result content blocks to [Tool Result: id] text (#527)
- Previously, tool_result blocks in user messages were silently dropped
- This caused an infinite loop when Claude Code + superpowers routed to Codex:
Codex never received the tool response and kept re-requesting the tool
- Now: tool_result → text block '[Tool Result: {id}]\n{content}'
- Handles string, array-of-text, and JSON-serialized content types
docs(issues): add Turbopack postinstall workaround on #509 and #508
docs(issues): note that #464 (API key provisioning) is on the v3.0 roadmap
upstreamErrorResponse() now guards against parsed.error being an
object (e.g. ElevenLabs { error: { message, status_code } }) instead
of blindly using it as the error message string.
Both audioSpeech.ts and audioTranscription.ts fixed.
- Add resolveAudioContentType() to map video/* MIME to audio/* (fixes .mp4 uploads returning 'no speech detected')
- Add detect_language=true for Deepgram auto-language detection (fixes non-English audio)
- Add punctuate=true for better output quality
- Forward language form param to Deepgram when provided
- Apply same Content-Type fix to HuggingFace handler
- Add preserveDeveloperRole option and model compat override
- Normalize developer→system in roleNormalizer when not preserving
- Translator runs normalizeRoles for Responses API with option
- UI: ModelCompatPopover with do not preserve developer toggle
- Add ZWS_README_V2 documenting cause and fix
Made-with: Cursor
Previously resolveModelAlias() output was used only for getModelTargetFormat()
but the original model was sent in translatedBody.model and to the executor.
Now effectiveModel is propagated to all downstream operations.
Search Playground (Phase 1):
- Web Search as 10th endpoint in Playground with isolated SearchPlayground component
- Endpoint selector moved first; Provider/Model/Send hidden when search selected
- Provider dropdown via GET /api/search/providers, formatted results with cache indicator
Search Tools page (Phase 2) at /dashboard/search-tools:
- Split panel: SearchForm (left) with query, provider, filters + ResultsPanel (right)
- Compare Providers: parallel queries with latency, cost, response size, URL overlap
- Rerank Pipeline: model selector from /v1/models, results with position delta
- Search History: last 10 searches from call_logs with replay
- Sidebar entry under Debug section
Backend:
- GET /api/search/providers — list providers with auth guard + SEARCH_CREDENTIAL_FALLBACKS
- GET /api/search/stats — cache stats, provider aggregates, recent searches (auth guard)
- Add local provider_nodes routing for /v1/rerank (oMLX, vLLM support)
Bug fixes (from F-27 PR #432):
- Fix Brave news normalizer: data.results directly, not data.news.results
- Enforce max_results truncation after normalization for all providers
- Fix EndpointPageClient: use /api/search/providers instead of /api/v1/search
- Add isAuthenticated() guards on /api/search/providers and /api/search/stats
Response size metric in results meta bar and compare table.
i18n: 30+ keys in search namespace (en.json)
Includes version bump — v2.6.9 — committed ATOMICALLY with all changes:
fixes:
- fix(ci/t11): Remove 'any' from comments in openai-responses.ts + chatCore.ts
(\bany\b regex counted comment text as explicit any violations)
- fix(chatCore/#409): Normalize unsupported content part types before forwarding
Cursor sends {type:'file'} for .md attachments; Copilot/OpenAI providers reject
with 'type has to be either image_url or text'. Now: file/document→text block,
unknown types dropped with debug log. Fixes claude-* models via github-copilot.
workflow:
- chore(generate-release): ATOMIC COMMIT RULE — npm version patch MUST run before
feature commits so the release tag always points to a commit with full changes
Audio endpoints (/v1/audio/speech and /v1/audio/transcriptions) only
supported hardcoded providers from audioRegistry.ts. Local inference
backends configured as provider_nodes (e.g., MLX-Audio, oMLX) could
not serve audio through OmniRoute.
This adds a Phase 3 fallback in the audio model parser that consults
provider_nodes from the database. Local providers with api_type=openai
are automatically available for audio routing via their prefix
(e.g., mlx-audio/tts-model, omlx/whisper-large-v3-turbo).
Design: injection pattern — Next.js route handlers load provider_nodes
(async DB query) and pass them to the sync parser as a parameter.
No cross-workspace imports, no breaking changes to existing parsers.
Changes:
- Add buildDynamicAudioProvider() in audioRegistry.ts
- Add Phase 3 (provider_nodes prefix match) to parseAudioModel()
- Extend parseSpeechModel/parseTranscriptionModel with optional
dynamicProviders parameter (backward compatible)
- Load and inject provider_nodes in speech/transcription route handlers
- Dynamic providers use authType=none (local, no credentials needed)
Anthropic API rejects requests containing {"type":"text","text":""} with
400 "text content blocks must be non-empty". Some clients like LiteLLM
passthrough and @ai-sdk/anthropic may forward empty text blocks as-is.
Filter out empty text content blocks from messages before calling
translateRequest, similar to how empty-name tools are already stripped.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Embedding endpoint (/v1/embeddings) only supports 6 hardcoded cloud
providers. Local inference backends (oMLX, Ollama) serving embeddings
via provider_nodes are inaccessible through OmniRoute.
This adds dynamic provider_node support for embeddings:
- Add EmbeddingProvider interface and buildDynamicEmbeddingProvider()
- Add Phase 2 (provider_nodes prefix match) in parseEmbeddingModel()
- Handler accepts resolvedProvider/resolvedModel from route (injection pattern)
- Handler supports authType=none for local providers (was missing — critical gap)
- Route loads local provider_nodes (localhost only — prevents auth bypass/SSRF)
- Route filters by apiType=chat|responses and localhost hostname
- buildDynamicEmbeddingProvider validates inputs (prefix + baseUrl required)
- Per-node try/catch in map — one bad row doesn't block all providers
- DB errors logged and fall back to hardcoded providers
Use shallow copy ({ ...body }) instead of direct reference assignment
so that later translatedBody.model = model does not mutate the
caller's original body object.
When both source and target formats are Claude, skip all request
modification and forward the body untouched. This prevents
prepareClaudeRequest from corrupting valid Claude-native requests
destined for anthropic-compatible provider nodes.
Reasoning models (o1, o1-pro, o3, o3-mini) reject standard parameters
like temperature and top_p with 400 Bad Request. OmniRoute's default
executor forwards all parameters without filtering.
This fix adds declarative parameter filtering:
- Add unsupportedParams[] field to RegistryModel interface
- Add REASONING_UNSUPPORTED frozen constant shared across entries
- Add o1-pro, o3, o3-mini to OpenAI registry (were missing)
- Add getUnsupportedParams() helper with:
- O(1) precomputed map lookup (not O(N×M) scan)
- Cross-provider routing support via precomputed map
- Prefixed model ID support (e.g., "openai/o3" → "o3")
- Strip unsupported params in chatCore.ts before executor call
- Use Object.hasOwn() for safe property check (no prototype chain)
- Log stripped params at WARN level for visibility
The filter introduced in #346 only checked OpenAI-format tool names
(tool.function.name), silently dropping all tools when the request
arrives in Anthropic Messages API format (tool.name without .function).
This happens when LiteLLM proxies requests with anthropic/ model prefix —
it translates to Anthropic format before forwarding, so OmniRoute receives
Claude-format tools. The filter drops them all, causing Anthropic API to
return 400: 'tool_choice.any may only be specified while providing tools'.
Fix: check both formats with fn?.name ?? tool.name.
All tests pass except pre-existing clearAccountError module resolution (dataPaths) which is unrelated to this PR. Merging codex native passthrough fix.
