Trigger Authoring Chat Agent

Authoring a chat agent

A chat.agent runs an entire conversation as one long-lived Trigger.dev task. It wakes when a message arrives, freezes when none do, and in-memory state survives page refreshes, deploys, idle gaps, and crashes. Your code is the loop you would write anyway: messages in, streamText out. There are no API routes. The frontend talks to the agent through a TriggerChatTransport, so history accumulates server-side and the client ships only the new message each turn.

Works with Vercel AI SDK v5, v6, or v7. On v7 also install @ai-sdk/otel so model calls are traced (the SDK registers it for you).

Setup

Three pieces: the agent task, two server actions, and the frontend transport.

1. Define the agent

import { chat } from "@trigger.dev/sdk/ai";
import { streamText, stepCountIs } from "ai";
import { anthropic } from "@ai-sdk/anthropic";

export const myChat = chat.agent({
  id: "my-chat",
  run: async ({ messages, signal

run receives messages already converted to ModelMessage[] (the SDK converts the frontend’s UIMessage[] for you) plus a signal that aborts on stop or cancel. Returning the StreamTextResult auto-pipes it to the frontend.

2. Add two server actions

Both run on your server, so the browser never holds your environment secret key. This is also where per-user / per-plan authorization and any paired DB writes live.

"use server";
import { auth } from "@trigger.dev/sdk";
import { chat } from "@trigger.dev/sdk/ai";

// Creates the Session + first run, returns a session PAT. Idempotent on (env, chatId).
export const startChatSession = chat.createStartSessionAction("my-chat");

// Pure mint. The transport calls this on 401/403 to refresh an expired token.
export async function mintChatAccessToken

3. Wire the frontend

"use client";
import { useState } from "react";
import { useChat } from "@ai-sdk/react";
import { useTriggerChatTransport } from "@trigger.dev/sdk/chat/react";
import type { myChat } from "@/trigger/chat";
import { mintChatAccessToken, startChatSession } from

The transport is memoized (created once, reused across renders). Passing typeof myChat flows the agent’s message type through useChat.

Core patterns

1. Return vs pipe

Return the streamText result from run for the simple case. When streamText is called deep inside nested helpers, call await chat.pipe(result) from anywhere in the task instead, and let run resolve void.

export const agentChat = chat.agent({
  id: "agent-chat",
  run: async ({ messages }) => {
    await runAgentLoop(messages); // don't return; pipe inside
  },
});

async function runAgentLoop(messages: ModelMessage[]) {

2. Typed tools (declare on config AND spread back)

Declare tools on chat.agent({ tools }), read them back typed from the run() payload, and pass that set to chat.toStreamTextOptions({ tools }). One declaration flows everywhere.

import { tool, stepCountIs } from "ai";
import { z } from "zod";

const tools = {
  searchDocs: tool({
    description: "Search the docs.",
    inputSchema: z.object({ query: z.string() }),
    execute: async ({ 

tools also accepts a function (event) => ToolSet resolved per turn, where event carries chatId, turn, continuation, and clientData.

3. Custom data parts (persisted vs transient)

data-* parts written via chat.response.write() in run() (or writer.write() in hooks) persist into responseMessage.parts and surface in onTurnComplete. Add transient: true to stream them without persisting. Writes via chat.stream are always ephemeral.

// In run() - persists, surfaces in onTurnComplete's responseMessage
chat.response.write({ type: "data-context", data: { searchResults } });

// In a hook via writer - streams but does NOT persist
writer.write({ type: "data-progress", id: "search", data: { percent: 50 }, transient: true });

4. Custom UIMessage type, client data, and builder hooks

For typed data-* parts or a tool map, build the agent through chat.withUIMessage<T>() and chat.withClientData({ schema }). Builder methods chain in any order; builder hooks run before the matching task hook. streamOptions becomes the default uiMessageStreamOptions (shallow-merged, agent wins).

export const myChat = chat
  .withUIMessage<MyChatUIMessage>({ streamOptions: { sendReasoning: true } })
  .withClientData({ schema: z.object({ userId: z.string() }) })
  .agent({
    id: "my-chat",
    tools: myTools,
    onTurnStart: async ({ uiMessages

Build MyChatUIMessage as UIMessage<unknown, MyDataTypes, InferUITools<typeof tools>> (or, for tools only, InferChatUIMessageFromTools<typeof tools> from @trigger.dev/sdk/ai). On the frontend, narrow useChat with InferChatUIMessage<typeof myChat> from @trigger.dev/sdk/chat/react.

5. Lifecycle hooks and stop

chat.agent accepts hooks that fire in a fixed per-turn order:

onValidateMessages -> hydrateMessages -> onChatStart (chat's first message only)
  -> onTurnStart -> run() -> onBeforeTurnComplete -> onTurnComplete

onBoot fires once per worker process (every fresh boot, including continuation runs) and is where chat.local, DB connections, and per-process state belong. onChatStart fires only on the chat’s first message. Suspend/resume use onChatSuspend / onChatResume. Config options include tools, clientDataSchema, maxTurns (100), turnTimeout (“1h”), idleTimeoutInSeconds (30), uiMessageStreamOptions, and exitAfterPreloadIdle. There is no generic retry; chat.agent runs with maxAttempts: 1 internally.

Stop is load-bearing: the signal passed to run aborts on stop or cancel. Forward it as abortSignal to streamText, or the Stop button updates the UI while the model keeps generating server-side.

run: async ({ messages, signal }) =>
  streamText({ ...chat.toStreamTextOptions(), model, messages, abortSignal: signal, stopWhen: stepCountIs(15) });

6. Migrating from a plain AI SDK streamText route

There is no API route in this model. The transport replaces the route round-trip, so:

• Delete the route handler. Move per-request auth into the two server actions from Setup step 2.
• Move the streamText call into run. It already receives pre-converted ModelMessage[].
• Return the StreamTextResult (it auto-pipes) and add ...chat.toStreamTextOptions() first.
• On the client, swap the api URL for useTriggerChatTransport; useChat stays the same shape.

Common mistakes

• CRITICAL: forgetting ...chat.toStreamTextOptions().

  tsCopy

  // Wrong - compaction / steering / background injection silently no-op
  return streamText({ model, messages, abortSignal: signal });
  // Correct - spread FIRST so explicit overrides win
  return streamText({ ...chat.toStreamTextOptions(), model, messages, abortSignal: signal });

  It wires the prepareStep callback behind compaction, mid-turn steering, and background injection, injects the system prompt from chat.prompt(), resolves the registry model, and adds telemetry. Omitting it makes all of those silently no-op with no error.

• Declaring tools only on streamText. Also declare them on chat.agent({ tools }), read them back from run, and pass chat.toStreamTextOptions({ tools }). Otherwise each tool’s toModelOutput runs on turn 1 but is dropped when history is re-converted on later turns.

• Not forwarding signal for stop. Without abortSignal: signal, Stop updates the UI but the model keeps generating server-side.

• Initializing chat.local in onChatStart. Initialize it in onBoot. onChatStart fires once per chat, so continuation runs skip it and crash with chat.local can only be modified after initialization. fires on every fresh worker.

References

• trigger-chat-agent-advanced skill - lifecycle hooks in depth, sessions, raw-task primitives (chat.createSession, chat.customAgent, chat.stream), compaction, HITL approvals, recovery.
• trigger-realtime skill - Realtime hooks and frontend streaming beyond the chat transport.
• trigger-tasks skill - base task() semantics, ctx, and standard lifecycle hooks.

Reference docs ship beside this skill in the same package, read them locally (no network), pinned to your installed version. The sources: frontmatter above lists every doc this skill draws from, all under @trigger.dev/sdk/docs/ai-chat/. Start with quick-start.mdx, backend.mdx, tools.mdx, types.mdx, frontend.mdx.

A chat.agent is a Trigger.dev task, so it builds and deploys like any other. For trigger.config.ts and build extensions (Prisma, Playwright, Python, FFmpeg, etc. — e.g. when a tool needs them), read the bundled config docs under @trigger.dev/sdk/docs/config/ (extensions are in config/extensions/, starting with overview.mdx).

Version

This skill is bundled inside @trigger.dev/sdk and read directly from node_modules, so it always matches your installed SDK version (see the adjacent package.json). The full documentation for these APIs ships alongside it under @trigger.dev/sdk/docs/.