Conversation
|
The latest updates on your projects. Learn more about Vercel for GitHub.
|
![chatgpt-codex-connector[bot]](https://avatars.githubusercontent.com/in/1144995?s=60&v=4){kind=small}
There was a problem hiding this comment.
💡 Codex Review
Here are some automated review suggestions for this pull request.
Reviewed commit: c84fa59791
ℹ️ About Codex in GitHub
Your team has set up Codex to 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 👍.
Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".
| for await (const textPart of textStream) { | ||
| yield { textPart }; |
There was a problem hiding this comment.
agentMode defaults to true and line 59 passes tools into streamLLM(), but this loop treats every streamed chunk as plain text; when the model emits a tool event, it gets forwarded as textPart: { ... } instead of a structured tool event. In the create-document flow this is concatenated into the suggestion string, producing invalid output like [object Object] whenever a tool is invoked. Please branch on event type here (as done in ChatThread streaming) or disable tools for this endpoint.
Useful? React with 👍 / 👎.
| return client.post('', { action: 'ChatThread.toggleAgentMode', ...params }).then(res => res.data); | ||
| }, | ||
| streamChatMessage: async function* streamChatMessage(params) { | ||
| // Don't stream on Next.js or Netlify for now. | ||
| const data = await client.post('', { action: 'ChatThread.createChatMessage', ...params }).then(res => res.data); |
There was a problem hiding this comment.
In the Lambda API branch, this commit adds toggleAgentMode, but chat requests still go through ChatThread.createChatMessage (line 70), which does not use chatThread.agentMode or tool-enabled prompting. So the new toggle appears to work but has no effect in Next.js/Netlify deployments. Either route Lambda chat through the agent-aware path or hide/disable this toggle for Lambda mode.
Useful? React with 👍 / 👎.
There was a problem hiding this comment.
Pull request overview
Adds an “Agent Mode” to chat threads that enables the LLM to call server-side tools (MongoDB exploration + script type-checking), streams tool-call status to the UI, and persists agent-mode + tool-call metadata.
Changes:
- Add
agentModeto chat threads, plus a new API/action to toggle it from the frontend. - Extend LLM streaming to emit tool-call/tool-result events and display tool activity in the chat UI.
- Introduce server-side agent tools (
estimatedDocumentCount,find,findOne,typeCheck) and persist tool-call history on assistant messages.
Reviewed changes
Copilot reviewed 16 out of 16 changed files in this pull request and generated 7 comments.
Show a summary per file
| File | Description |
|---|---|
| package.json | Adds runtime deps needed for agent tooling (notably TypeScript). |
| frontend/src/chat/chat.js | Adds tool-call handling during streaming + agent-mode toggle method + scroll helper. |
| frontend/src/chat/chat.html | Adds Agent Mode toggle button and minor layout/icon tweaks. |
| frontend/src/chat/chat-message/chat-message.js | Adds tool input formatting for tool-call display. |
| frontend/src/chat/chat-message/chat-message.html | Renders tool-call status list above assistant messages; adjusts message width. |
| frontend/src/chat/chat-message-script/chat-message-script.html | Minor styling adjustments to script editor UI controls. |
| frontend/src/api.js | Adds ChatThread.toggleAgentMode() client API method. |
| backend/integrations/streamLLM.js | Switches to fullStream and emits text/tool-call/tool-result events. |
| backend/helpers/getAgentTools.js | New agent tool definitions (MongoDB exploration + script type-check). |
| backend/db/chatThreadSchema.js | Persists agentMode on chat threads. |
| backend/db/chatMessageSchema.js | Persists toolCalls metadata on chat messages. |
| backend/authorize.js | Adds authorization mapping for ChatThread.toggleAgentMode. |
| backend/actions/Model/streamChatMessage.js | Enables agent tools for model chat streaming (currently impacts streaming contract). |
| backend/actions/ChatThread/toggleAgentMode.js | New action to toggle agentMode for a thread. |
| backend/actions/ChatThread/streamChatMessage.js | Uses agent system prompt + emits tool-call/tool-result events + stores toolCalls. |
| backend/actions/ChatThread/index.js | Exports the new toggleAgentMode action. |
Comments suppressed due to low confidence (1)
backend/actions/Model/streamChatMessage.js:64
Model.streamChatMessagenow passes through the output ofstreamLLM(), which can emit tool-call/tool-result events (objects). The current loop wraps every event as{ textPart: event }, so tool-call objects will be sent astextPartand break consumers that expecttextPartto be a string (e.g. create-document AI streaming). Update the loop to detecttypeof event === 'string'and yield{ textPart: event }, otherwise yield the tool event as-is or ignore it for this endpoint.
const llmMessages = [{ role: 'user', content: [{ type: 'text', text: content }] }];
const llmOptions = agentMode ? { ...options, tools: getAgentTools(db) } : options;
const textStream = streamLLM(llmMessages, system, llmOptions);
for await (const textPart of textStream) {
yield { textPart };
}
💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.
Co-authored-by: Copilot <175728472+Copilot@users.noreply.github.com>
|
Just reading the AI generated PR, i think merging this in would be helpful for the create document modal issue |
Co-authored-by: Copilot <175728472+Copilot@users.noreply.github.com>
…nto vkarpov15/agents-wip
No description provided.