Transcripts

Configure identity and transcripts together#

Transcripts are indexed by a user key — a stable, platform-independent identifier you derive from the inbound author. Without an identity resolver the bot can't map a Slack user to a Teams user (or even to their own previous Slack messages under a different thread), so identity and transcripts are configured as a pair:

import { createBot } from "@copilotkit/bot";
import { createRedisStore } from "@copilotkit/bot-store-redis";

const bot = createBot({
  adapters: [/* ... */],
  agent: (threadId) => /* ... */,
  store: {
    adapter: createRedisStore({ url: process.env.REDIS_URL! }),
    // Map any platform's author object to a stable user key.
    // Return null to opt a user out of transcript tracking.
    identity: ({ author }) => author.email ?? null,
    transcripts: {
      retention: "30d",      // auto-delete entries older than 30 days
      maxPerUser: 200,       // keep at most 200 entries per user key
    },
  },
});

The author object is provided by the platform adapter — on Slack it includes the user's Slack profile fields including email if your bot has the users:read.email scope. Use whatever field gives you a stable, cross-platform identifier — email is the most common choice.

The easy path — runAgent with transcript injection#

Pass transcript: true to thread.runAgent() and the bot handles everything automatically:

1. Fetches the user's prior messages (up to 50 entries by default, oldest-first).
2. Injects them into the agent's context, labelled by platform and timestamp.
3. Appends the current user turn.
4. Captures the streamed assistant reply and appends it to the transcript.

bot.onMention(async ({ thread }) => {
  await thread.runAgent({ transcript: true }); 
});

To control how many prior messages are injected, pass a limit instead:

await thread.runAgent({ transcript: { limit: 20 } });

What the agent sees — platform-labelled context#

The injected history is formatted so the agent knows which platform each message came from. A user who asked a question on Teams and is now following up on Slack produces context like:

[teams | 2026-06-20T09:14:22Z] User: Can you pull the Q2 numbers from the dashboard?
[teams | 2026-06-20T09:14:35Z] Assistant: Here are the Q2 figures: …
[slack  | 2026-06-22T14:03:10Z] User: Actually, can you do that again but filter to EMEA only?

The agent sees the full cross-platform trail and can reference it naturally — "You asked on Teams last week about Q2 numbers — here's the EMEA slice." Each entry carries both platform and ts so you can build custom UIs, audit logs, or export pipelines on top of the raw data.

Manual control — append, list, and delete#

For cases where runAgent doesn't fit (proactive messages, non-agent flows, GDPR tooling), the full API is on bot.transcripts:

Append a message manually:

await bot.transcripts.append(thread, {
  role: "user",
  content: "Can you check on my open ticket?",
});
// Provide userKey explicitly if you're outside a request context:
await bot.transcripts.append(thread, msg, { userKey: "user@example.com" });

List a user's history (oldest-first, with filters):

const history = await bot.transcripts.list({
  userKey: "user@example.com",
  limit: 50,
  platforms: ["slack", "teams"],  // omit to include all platforms
  roles: ["user"],                // omit to include assistant turns too
  threadId: thread.id,            // omit to include all threads
});

for (const entry of history) {
console.log(`[${entry.platform} | ${entry.ts}] ${entry.role}: ${entry.content}`);
}

Delete all history for a user (GDPR right-to-erasure):

const result = await bot.transcripts.delete({ userKey: "user@example.com" });
console.log(`Deleted ${result.deleted} entries`);