Chat
Chat lets you ask natural language questions about your knowledge graph and documents. The AI retrieves relevant context from your project and generates a grounded response with citations.
Starting a Conversation
Section titled “Starting a Conversation”Click Chat in the sidebar and start typing.
POST /api/chat/conversationsContent-Type: application/json
{ "title": "Database architecture questions", "projectId": "proj_xyz789"}Sending a Message
Section titled “Sending a Message”Returns a server-sent event (SSE) stream with partial tokens as they are generated:
POST /api/chat/streamContent-Type: application/json
{ "conversationId": "<uuid>", "message": "What decisions did we make about the database?"}SSE events:
data: {"type":"token","content":"We decided "}data: {"type":"token","content":"to use PostgreSQL "}data: {"type":"citations","citations":[{"objectId":"...","title":"Use PostgreSQL"}]}data: {"type":"done"}POST /api/chat/{conversationId}/messagesContent-Type: application/json
{ "role": "user", "content": "What decisions did we make about the database?"}Conversations
Section titled “Conversations”List conversations
Section titled “List conversations”GET /api/chat/conversationsGet a conversation with messages
Section titled “Get a conversation with messages”GET /api/chat/{id}Update a conversation
Section titled “Update a conversation”PATCH /api/chat/{id}Content-Type: application/json
{ "title": "Renamed conversation", "draftText": "Work in progress..."}Delete a conversation
Section titled “Delete a conversation”DELETE /api/chat/{id}Conversation fields
Section titled “Conversation fields”| Field | Description |
|---|---|
title |
Display name for the conversation |
isPrivate |
If true, only visible to you (default: true) |
enabledTools |
List of tool names the AI can use in this conversation |
agentDefinitionId |
Optional: link to a specific agent definition to use as the AI persona |
objectId / canonicalId |
Optional: pin the conversation to a specific graph object for focused Q&A |
Citations
Section titled “Citations”Every assistant response includes citations — references to the graph objects or document chunks used to generate the answer. Citations appear in the citations field of the message:
{ "role": "assistant", "content": "We decided to use PostgreSQL because of team expertise.", "citations": [ { "type": "graph_object", "objectId": "<canonical_id>", "title": "Use PostgreSQL", "snippet": "Team expertise and existing infrastructure" } ]}The chat AI can call tools to answer questions. The available tools depend on what is enabled for the conversation:
| Tool | What it does |
|---|---|
graph_query |
Searches and retrieves graph objects |
search |
Runs hybrid search across documents and objects |
graph_create_object |
Creates a new graph object (write-enabled conversations) |
Control which tools are available via enabledTools when creating or updating a conversation.
- Be specific about type: “What
Decisionobjects relate to authentication?” gives more precise results than “What do we know about auth?”. - Reference objects by name: The AI can look up objects by their title or key property.
- Long conversations: The platform maintains a context summary as conversations grow to keep responses relevant.