β¨ feat: add GLM-5 model support to LobeHub provider#13189
Merged
Conversation
|
The latest updates on your projects. Learn more about Vercel for GitHub.
|
Member
Codecov Reportβ Patch coverage is Additional details and impacted files@@ Coverage Diff @@
## canary #13189 +/- ##
===========================================
- Coverage 84.35% 74.21% -10.15%
===========================================
Files 375 1535 +1160
Lines 24447 126184 +101737
Branches 4740 13913 +9173
===========================================
+ Hits 20623 93643 +73020
- Misses 3731 32430 +28699
- Partials 93 111 +18
Flags with carried forward coverage won't be shown. Click here to find out more.
π New features to boost your workflow:
|
There was a problem hiding this comment.
π‘ Codex Review
Here are some automated review suggestions for this pull request.
Reviewed commit: 3ea55e504f
βΉοΈ About Codex in GitHub
Codex has been enabled to automatically review pull requests in this repo. Reviews are triggered when you
- Open a pull request for review
- Mark a draft as ready
- Comment "@codex review".
If Codex has suggestions, it will comment; otherwise it will react with π.
When you sign up for Codex through ChatGPT, Codex can also answer questions or update the PR, like "@codex address that feedback".
Comment on lines
+45
to
+48
| arguments: z.string().catch(''), | ||
| name: z.string().catch(''), | ||
| }), | ||
| id: z.string(), | ||
| id: z.string().catch(''), |
There was a problem hiding this comment.
Don't swallow missing tool ids/names in streamed tool calls
For the GLM-5 case mentioned in the comment, these fallbacks do not let later chunks βfill in the correct valuesβ. OpenAIStream still assigns a synthetic id to the first partial chunk, and parseToolCalls() (packages/model-runtime/src/helpers/parseToolCalls.ts) only concatenates function.arguments; it never backfills function.name/id, and if a later chunk finally includes the real id it is treated as a second tool call. In practice, onToolsCalling consumers will see an empty or duplicated tool during streaming for Zhipu/GLM-5 instead of a single progressively completed call.
Useful? React with πΒ / π.
| if (reader) { | ||
| let done = false; | ||
| while (!done) { | ||
| const { value, done: isDone } = await reader.read(); |
- Add GLM-5 model card with USD fixed pricing - Register zhipu as a new ApiType in RouterRuntime - Pass payload through handleStream for correct billing
Some providers (e.g., Zhipu GLM-5) send null for function.name/id in early streaming tool_call chunks. Use .catch() to prevent ZodError from breaking the SSE stream.
When a model calls tools, the stop chunk may not reach handleChunk, leaving the reasoning spinner stuck. Call endReasoningIfNeeded() at the start of handleFinish to ensure reasoning always completes.
β¦ary reasoning guard - Add test verifying payload is passed to custom handleStream - Revert endReasoningIfNeeded guard in handleFinish (reasoning is already ended by handleToolCallsChunk and handleTextChunk)
The .catch('') on MessageToolCallSchema was only needed for aihubmix
proxy which sends null values; the official zai API sends correct data.
Remove the global schema relaxation and its associated tests.
With tool_stream enabled, some proxies (e.g., aihubmix) send empty placeholder tool_call chunks without id/function.name before the real data arrives. Filter them out to prevent ZodError in parseToolCalls.
fd15036 to
84ee107
Compare
Member
|
β€οΈ Great PR @tjx666 β€οΈ The growth of project is inseparable from user feedback and contribution, thanks for your contribution! If you are interesting with the lobehub developer community, please join our discord and then dm @arvinxx or @canisminor1990. They will invite you to our private developer channel. We are talking about the lobe-chat development or sharing ai newsletter around the world. |
hardy-one
pushed a commit
to hardy-one/lobehub
that referenced
this pull request
Mar 23, 2026
ONLY-yours
added a commit
that referenced
this pull request
Mar 27, 2026
# π release: 20260326 This release includes **91 commits**. Key updates are below. - **Agent can now execute background tasks** β Agents can perform long-running operations without blocking your conversation. [#13289](#13289) - **Better error messages** β Redesigned error UI across chat and image generation with clearer explanations and recovery options. [#13302](#13302) - **Smoother topic switching** β No more full page reloads when switching topics while an agent is responding. [#13309](#13309) - **Faster image uploads** β Large images are now automatically compressed to 1920px before upload, reducing wait times. [#13224](#13224) - **Improved knowledge base** β Documents are now properly parsed before chunking, improving retrieval accuracy. [#13221](#13221) ### Bot Platform - **WeChat Bot support** β You can now connect LobeChat to WeChat, in addition to Discord. [#13191](#13191) - **Richer bot responses** β Bots now support custom markdown rendering and context injection. [#13294](#13294) - **New bot commands** β Added `/new` to start fresh conversations and `/stop` to halt generation. [#13194](#13194) - **Discord stability fixes** β Fixed thread creation issues and Redis connection drops. [#13228](#13228) [#13205](#13205) ### Models & Providers - **GLM-5** is now available in the LobeHub model list. [#13189](#13189) - **Coding Plan providers** β Added support for code planning assistant providers. [#13203](#13203) - **Tencent Hunyuan 3.0 ImageGen** β New image generation model from Tencent. [#13166](#13166) - **Gemini content handling** β Better handling when Gemini blocks content due to safety filters. [#13270](#13270) - **Claude token limits fixed** β Corrected max window tokens for Anthropic Claude models. [#13206](#13206) ### Skills & Tools - **Auto credential injection** β Skills can now automatically request and use required credentials. [#13124](#13124) - **Smarter tool permissions** β Built-in tools skip confirmation for safe paths like `/tmp`. [#13232](#13232) - **Model switcher improvements** β Quick access to provider settings and visual highlight for default model. [#13220](#13220) ### Memory - **Bulk delete memories** β You can now delete all memory entries at once. [#13161](#13161) - **Per-agent memory control** β Memory injection now respects individual agent settings. [#13265](#13265) ### Desktop App - **Gateway connection** β Desktop app can now connect to LobeHub Gateway for enhanced features. [#13234](#13234) - **Connection status indicator** β See gateway connection status in the titlebar. [#13260](#13260) - **Settings persistence** β Gateway toggle state now persists across app restarts. [#13300](#13300) ### CLI - **API key authentication** β CLI now supports API key auth for programmatic access. [#13190](#13190) - **Shell completion** β Tab completion for bash/zsh/fish shells. [#13164](#13164) - **Man pages** β Built-in manual pages for CLI commands. [#13200](#13200) ### Security - **XSS protection** β Sanitized search result image titles to prevent script injection. [#13303](#13303) - **Workflow hardening** β Fixed potential shell injection in release automation. [#13319](#13319) - **Dependency update** β Updated nodemailer to address security advisory. [#13326](#13326) ### Bug Fixes - Fixed skill page not redirecting correctly after import. [#13255](#13255) [#13261](#13261) - Fixed token counting in group chats. [#13247](#13247) - Fixed editor not resetting when switching to empty pages. [#13229](#13229) - Fixed manual tool toggle not working. [#13218](#13218) - Fixed Search1API response parsing. [#13207](#13207) [#13208](#13208) - Fixed mobile topic menus rendering issues. [#12477](#12477) - Fixed history count calculation for accurate context. [#13051](#13051) - Added missing Turkish translations. [#13196](#13196) ### Credits Huge thanks to these contributors: @bakiburakogun @hardy-one @Zhouguanyang @sxjeru @hezhijie0327 @arvinxx @cy948 @CanisMinor @Innei @lijian @lobehubbot @neko @rdmclin2 @rivertwilight @tjx666
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
2 participants
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
π» Change Type
π Related Issue
Fixes LOBE-5086
π Description of Change
Add GLM-5 model support to the LobeHub hosted provider, along with two streaming bug fixes discovered during testing:
Model card & runtime
lobehub/chat/zhipu.ts) with USD fixed pricing ($1/M input, $3.2/M output)zhipuas a newApiTypein RouterRuntime withLobeZhipuAImappingpayloadthroughhandleStreamso pricing flows towithUsageCostfor correct billingBug fixes
MessageToolCallSchema: use.catch()on all fields to prevent ZodError when providers (e.g., GLM-5) send null in early streaming tool_call chunksStreamingHandler.handleFinish: callendReasoningIfNeeded()to ensure the reasoning spinner completes even when the stop chunk doesn't reachhandleChunk(tool call scenarios)π§ͺ How to Test
StreamingHandler.test.ts: 2 new test cases for handleFinish reasoning guardTest prompts used with GLM-5: