* feat(qoder): native cosy integration
* feat(qoder): implement native COSY encryption algorithm and remove CLI child instances, plus workflow bumps
* feat(resilience): context overflow fallback, OAuth token detection, empty content guard & context-optimized combo strategy
- Add isContextOverflowError + isContextOverflow detectors (400 + token-limit signals)
- Auto-fallback to next family model on context overflow in chatCore
- Add isEmptyContentResponse to catch fake-success empty responses, trigger fallback + recursive retry
- Add OAUTH_INVALID_TOKEN error type (T11) with isOAuthInvalidToken signal matching; warn instead of deactivating node
- Add getModelContextLimit helper in modelsDevSync (reads limit_context from synced capabilities)
- Upgrade getTokenLimit in contextManager to check models.dev DB before registry (fixes gemini-2.5-pro: 1000000→1048576)
- Add findLargerContextModel in modelFamilyFallback for context-aware model selection
- Add sortModelsByContextSize + context-optimized combo strategy in combo.ts
- Update context-manager unit test for corrected gemini-2.5-pro limit
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
* fix(review): address Gemini code review — tool_calls path, infinite recursion, dedup signals, findLargerContextModel
- Fix isEmptyContentResponse: check message.tool_calls/delta.tool_calls instead
of firstChoice.tool_calls (wrong OpenAI API path, caused tool-call responses
to be falsely flagged as empty)
- Fix empty content fallback: replace recursive handleChatCore call (infinite
recursion risk + wrong model due to original body.model) with non-recursive
pattern — call executeProviderRequest, parse fallback response body, reassign
responseBody and fall through to existing processing
- Fix context overflow: use findLargerContextModel over family candidates first,
fall back to getNextFamilyFallback — ensures we pick a model with actually
larger context window on overflow
- Fix signal dedup: export CONTEXT_OVERFLOW_SIGNALS + CONTEXT_OVERFLOW_REGEX
from errorClassifier.ts; import shared regex in modelFamilyFallback.ts,
removing duplicate signal list and per-call RegExp construction
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
* fix(UI): add context-optimized strategy to frontend schema and options
* fix(sse): preserve Responses API events in stream translation
When translating Claude-format responses (e.g. GLM) to Responses API
format for Codex CLI, the sanitizer stripped {event, data} structured
items to {"object":"chat.completion.chunk"}, losing all content and
the critical response.completed event.
Only run sanitizeStreamingChunk on OpenAI Chat Completions chunks,
skipping items that have the Responses API {event, data} structure.
* test(sse): add regression test for Claude→Responses stream sanitization
Verifies that {event,data} structured items from the Responses API
translator bypass sanitizeStreamingChunk when translating Claude-format
providers (e.g. GLM) to Responses API format for Codex CLI.
* fix(sse): strengthen Responses API event detection with response. prefix check
Use explicit `response.` prefix check instead of generic `event && data`
presence check, as recommended in PR review.
* fix: pin Next.js to 16.0.10 to prevent Turbopack hashed module bug
Remove ^ prefix from next and eslint-config-next to prevent
automatic upgrades to 16.1.x+ which introduced content-based
hashing for external module references in Turbopack.
Also remove duplicate Material Symbols @import from globals.css
(font already loaded via <link> in layout.tsx).
Fixes#509
* align cc-compatible cache handling with client passthrough
* chore: integrate resilience and turbopack fixes (PRs #992, #990, #987)
* chore(release): bump to v3.5.2 — changelog, docs, version sync
* docs(i18n): sync documentation updates to 33 languages
* fix(qoder): replace any with unknown to comply with strict any-budget
---------
Co-authored-by: diegosouzapw <diegosouzapw@users.noreply.github.com>
Co-authored-by: oyi77 <oyi77@users.noreply.github.com>
Co-authored-by: Claude Sonnet 4.6 <noreply@anthropic.com>
Co-authored-by: Chris Staley <christopher-s@users.noreply.github.com>
Co-authored-by: Ivan <shanin-i2011@yandex.ru>
Co-authored-by: R.D. <rogerproself@gmail.com>
- Remove x-goog-user-project header and executor-level project override
that caused 403 "Cloud Code Private API has not been used in project X"
- Add PROJECT_ROUTE_ERROR classifier type so project-routing 403s don't
permanently ban accounts (keeps accounts active, tracks the error)
- Fix Cloud Code envelope unwrapping for content accumulation in stream.ts
(Cloud Code wraps responses in { response: { candidates: [...] } })
- Extract unwrapGeminiChunk() into streamHelpers.ts with format guard
- Remove _currentModel singleton race condition from GeminiCLIExecutor
- Add handler for PROJECT_ROUTE_ERROR in chatCore.ts
- Add TODO in antigravity.ts about same stale-project risk
- Add 7 unit tests for error classifier and stream unwrap paths
* fix(stream): normalize delta.reasoning to reasoning_content in SSE streaming
NVIDIA kimi-k2.5 (and potentially other providers) send reasoning
tokens as `delta.reasoning` in SSE streaming chunks instead of the
standard OpenAI `delta.reasoning_content` field. This caused reasoning
content to be silently dropped during stream passthrough — clients
received only the final answer with no reasoning separation.
The non-streaming sanitizer (responseSanitizer.ts) already handled this
alias, but the streaming pipeline did not.
Fix applied in 4 locations:
- stream.ts passthrough: normalize + force re-serialize sanitized chunk
- stream.ts translate: accumulate reasoning from delta.reasoning
- sseParser.ts: collect delta.reasoning in parseSSEToOpenAIResponse
- streamPayloadCollector.ts: collect delta.reasoning in buildOpenAISummary
* fix: eliminate injectedUsage reuse bug and add reasoning alias tests
- Detect delta.reasoning alias before sanitizeStreamingChunk() which
already normalizes it, removing dead post-sanitization normalization
- Replace injectedUsage reuse with separate needsReserialization flag
so reasoning re-serialization cannot block finish_reason/usage
mutations on the same SSE chunk (fixes CRITICAL review finding)
- Add unit test for parseSSEToOpenAIResponse reasoning alias
- Add unit test for buildStreamSummaryFromEvents reasoning alias
* fix(stream): separate reasoning from content in passthrough response body
The passthroughAccumulatedContent variable was mixing delta.content and
delta.reasoning_content into one string, causing the client_response
log and responseBody to lose reasoning separation.
- Add passthroughAccumulatedReasoning accumulator for reasoning deltas
- Set message.reasoning_content in responseBody when reasoning exists
- Only accumulate delta.content into passthroughAccumulatedContent
* fix: trim leading whitespace from assembled content in log summaries
NVIDIA and other providers emit token deltas with leading spaces
(e.g. ' The', ' user'). When joined, these produce a leading space in
the provider_response and parsed non-streaming response logs. Trim
the joined content and reasoning_content in both buildOpenAISummary
and parseSSEToOpenAIResponse for consistent log output.
* fix(stream): split combined reasoning+content deltas into separate SSE events
Some providers (e.g. NVIDIA NIM) send transition chunks with both
`delta.reasoning` and `delta.content` in the same SSE event.
After sanitization this becomes `reasoning_content` + `content`,
which violates the standard OpenAI streaming contract where these
fields are never mixed. Clients using if/else logic (LobeChat, etc.)
skip content when reasoning_content is present, losing the first
content token.
Split such combined chunks into two separate SSE events:
1. Reasoning-only event (finish_reason=null, no usage)
2. Content-only event (carries finish_reason and usage)
- 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
Passthrough mode now detects Responses SSE payloads (parsed.type starts
with 'response.') and skips Chat Completions-specific sanitization:
- sanitizeStreamingChunk() only runs on Chat Completions payloads
- fixInvalidId() and hasValuableContent() checks skipped for Responses
- Usage extraction still runs for both formats
- Content length tracking adapted for Responses delta format
This prevents potential stream corruption when Responses SSE data
triggers idFixed or other Chat Completions-specific rewrite conditions.
OmniRoute is an intelligent API gateway that unifies 20+ AI providers behind a single
OpenAI-compatible endpoint. Features include intelligent routing with 6 strategies,
multi-format translation (OpenAI/Claude/Gemini/Responses API), circuit breakers,
semantic caching, combo fallback chains, real-time health monitoring, and a full
dashboard with provider management, analytics, and CLI tool integration.
Key highlights:
- 20+ providers (Claude Code, Codex, Gemini CLI, GitHub Copilot, iFlow, Qwen, Kiro, etc.)
- 6 routing strategies (Fill First, Round Robin, P2C, Random, Least Used, Cost Optimized)
- Export/Import database backup with full archive support
- Translator Playground with 4 modes (Playground, Chat Tester, Test Bench, Live Monitor)
- 100% TypeScript across src/ and open-sse/
- Docker support with multi-stage builds
- Comprehensive documentation and 9 dashboard screenshots
