Reviewed and approved via consolidated analysis. GLM-5.1 addition and pricing corrections match official Z.AI pricing page. All 5 files follow existing patterns.
Reviewed and approved via consolidated analysis. Fix is surgical (1 line removed) with 122 lines of regression tests covering stream=true, stream=false and guard scenarios. Resolves#677.
Three tests covering the fixed bug where translateRequest received an
object instead of a boolean for the stream parameter:
- stream=true round-trip produces boolean true
- stream=false round-trip produces boolean false
- guard test documenting that passing an object as stream breaks typing
Co-Authored-By: Craft Agent <agents-noreply@craft.do>
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
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>
- Add glm-5.1 model to GLM Coding provider with fitness scores
- Update glm-5 pricing to match Z.AI API page ($1/$3.2/$0.2)
- Set glm-5.1 pricing to $1.2/$5/$0.3 per Z.AI
- Remove glm-4-32b (deprecated, returns empty from upstream)
- Rename Z.AI provider display name from "Z.AI (GLM-5)" to "Z.AI"
- Update zai pricing section to match glm pricing
In OpenAI Chat Completions streaming format, the tool call id and type
should only appear on the first chunk (tool declaration). Subsequent
argument delta chunks should only include index and function.arguments.
Including id on every delta chunk caused openai-to-claude.ts to emit
a new content_block_start for each chunk, breaking Claude Code ACP
sessions with malformed Claude-format streams.
Fixes#682
Co-authored-by: Chris Staley <christopher.staley@protonmail.com>
The hasValuableContent() function in streamHelpers.ts returned undefined
instead of explicit false when checking empty delta chunks. This caused
JavaScript type coercion issues where undefined !== '' evaluated to true,
passing empty chunks through to clients.
Fix: Replace implicit returns with explicit boolean returns using
typeof checks and length comparisons for all content fields (content,
reasoning_content, tool_calls, text, thinking, partial_json).
Test: Added unit tests covering OpenAI, Claude, and Gemini format edge cases.
Co-authored-by: oyi77 <oyi77@users.noreply.github.com>
Changes:
- fix: restore native Claude tool names in passthrough responses (PR #663 by @coobabm)
- fix: Clear All Models button now also removes aliases (PR #664 by @rdself)
- fix: completed truncated test from PR #663, added Claude-to-Claude passthrough test
- docs: update CHANGELOG and OpenAPI spec
The Clear All Models button was only deleting custom models from the
database but leaving their aliases intact, so the UI didn't reflect
the deletion. Now it also deletes all aliases belonging to the provider
and refreshes the alias state.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
High backoffLevel (up to 15) persisted permanently in the DB after a burst of 429s. The account health score dropped to zero (100 - 15*10 = -50), causing the account selector to never pick the account again. Only a successful request could reset backoffLevel via clearAccountError, but the account was never selected — creating a deadlock.
Now, during account selection, any non-terminal connection whose rateLimitedUntil has passed gets its backoffLevel reset to 0 and testStatus restored to active. The DB update is fire-and-forget to avoid blocking the hot path.
Ultraworked with [Sisyphus](https://github.com/code-yeongyu/oh-my-opencode)
Co-authored-by: Sisyphus <clio-agent@sisyphuslabs.ai>
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
- Fix Ollama Cloud base URL from api.ollama.com to ollama.com/v1/chat/completions
- Fix Ollama Cloud models URL to ollama.com/api/tags
- Add gemini-3.1-pro-preview and gemini-3.1-flash-lite-preview to Antigravity provider
Closes#643, closes#645
Thanks @brendandebeasi for this excellent contribution! 🎉 The bounded retry with exponential backoff is exactly the right approach for expired connections. Merged and will be included in v3.1.1.
Connections marked as 'expired' were permanently skipped by the health check scheduler (line 176: if testStatus === "expired" return). A single transient refresh failure could permanently disable auto-refresh, requiring manual re-authentication.
Replace the hard skip with a bounded retry mechanism: up to 3 attempts with exponential backoff (5min, 10min, 20min). On success, the connection is fully restored to active. On exhaustion, it remains expired (same as before). The existing circuit breaker (5 failures → 30min pause) provides additional protection.
Ultraworked with [Sisyphus](https://github.com/code-yeongyu/oh-my-opencode)
Co-authored-by: Sisyphus <clio-agent@sisyphuslabs.ai>
OpenAI-compatible clients (OpenCode, etc.) check capabilities/input_modalities fields on the /v1/models response to determine if a model supports image input. Omniroute was not emitting these fields, causing clients to assume text-only for all models routed through the proxy.
Add keyword-based vision detection (matching the existing playground heuristic) that annotates model entries with capabilities:{vision:true}, input_modalities:["text","image"], and output_modalities:["text"] for known multimodal models (GPT-4o/4-turbo, Claude 3+, Gemini, Pixtral, Qwen-VL, etc.).
Ultraworked with [Sisyphus](https://github.com/code-yeongyu/oh-my-opencode)
Co-authored-by: Sisyphus <clio-agent@sisyphuslabs.ai>
- 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.
- OpenAI-to-Claude translator now maps reasoning_effort (low/medium/high/max)
to Claude's thinking.budget_tokens. Fixes clients like OpenCode sending
reasoning_effort via @ai-sdk/openai-compatible losing thinking configuration.
- Ensures max_tokens > budget_tokens for all thinking configs.
- Token health check now proactively refreshes tokens within 5 min of expiry,
regardless of the configured health check interval — addresses Qwen OAuth
token refresh failures between scheduled checks.
