Skip to main content
A conversation keeps rolling context across turns so follow-ups like “how does that compare to Q3?” resolve correctly.

​
Start a session

const { session_id } = await client.startConversation();
The returned session_id is opaque — store it in your app state.

​
Continue a session

const turn1 = await client.continueConversation({
  query: "What was ACME's Q4 revenue?",
  sessionId: session_id,
});

const turn2 = await client.continueConversation({
  query: "How does that compare to Q3?",
  sessionId: session_id,
});
Under the hood this calls query() with maintainContext: true and the sessionId attached. You can do that yourself — continueConversation is just a named shortcut.

​
Stream a session turn

Use queryStream with the session ID: 
for await (const token of client.queryStream({
  query: "And what drove the increase?",
  sessionId: session_id,
  maintainContext: true,
})) {
  process.stdout.write(token);
}

​
List + inspect conversations

const conversations = await client.listConversations();
for (const c of conversations) {
  console.log(c.session_id, c.title ?? "(untitled)", c.turn_count);
}

const full = await client.getConversation(session_id);
console.log(full.messages);

​
Delete

await client.deleteConversation(session_id);
Deletes the session record and its rolling history. Source documents and facts are unaffected.

​
Scope a session to specific uploads

Useful for per-conversation knowledge isolation (e.g. customer-specific bots): 
const answer = await client.query({
  query: "What does this contract say about termination?",
  sessionId: session_id,
  maintainContext: true,
  conversationUploadIds: ["upload_customerA_contract"],
});
The retrieval restricts itself to those uploads for this session.