Back to recipes

Working with Workflows

Create and run durable workflows with steps, streaming, and agent execution. Covers starting, resuming, and persisting workflow results.

Skills

This recipe requires you to complete 1 recipe

Complete these prerequisites first

Workflow Development Kit Setup

Install and configure the Workflow Development Kit for resumable, durable AI agent workflows with step-level persistence, stream resumption, and agent orchestration.

Working with Workflows

Create and run durable workflows with steps, streaming, and agent execution. Covers starting, resuming, and persisting workflow results.

See:

  • Resource: using-workflows in Fullstack Recipes
  • URL: https://fullstackrecipes.com/recipes/using-workflows

Workflow Folder Structure

Each workflow has its own subfolder in src/workflows/:

src/workflows/
  steps/           # Shared step functions
    stream.ts      # UI message stream helpers
  chat/
    index.ts       # Workflow orchestration function ("use workflow")
    steps/         # Workflow-specific steps ("use step")
      history.ts
      logger.ts
      name-chat.ts
    types.ts       # Workflow-specific types
  • workflows/steps/ - Shared step functions reusable across workflows (e.g., stream helpers).
  • index.ts - Contains the main workflow function with the "use workflow" directive. Orchestrates the workflow by calling step functions.
  • steps/ - Contains individual step functions with the "use step" directive. Each step is a durable checkpoint.
  • types.ts - Type definitions for the workflow's UI messages.

Creating a Workflow

Define workflows with the "use workflow" directive:

ts
import { getWorkflowMetadata, getWritable } from "workflow";
import { startStream, finishStream } from "../steps/stream";
import { chatAgent } from "@/lib/ai/chat-agent";

export async function chatWorkflow({ chatId, userMessage }) {
  "use workflow";

  const { workflowRunId } = getWorkflowMetadata();

  // Persist user message
  await persistUserMessage({ chatId, message: userMessage });

  // Create assistant placeholder with runId for resumption
  const messageId = await createAssistantMessage({
    chatId,
    runId: workflowRunId,
  });

  // Get message history
  const history = await getMessageHistory(chatId);

  // Start the UI message stream
  await startStream(messageId);

  // Run agent with streaming
  const { parts } = await chatAgent.run(history, {
    maxSteps: 10,
    writable: getWritable(),
  });

  // Persist and finalize
  await persistMessageParts({ chatId, messageId, parts });

  // Finish the UI message stream
  await finishStream();

  await removeRunId(messageId);
}

Starting a Workflow

Use the start function from workflow/api:

typescript
import { start } from "workflow/api";
import { chatWorkflow } from "@/workflows/chat";

const run = await start(chatWorkflow, [{ chatId, userMessage }]);

// run.runId - unique identifier for this run
// run.readable - stream of UI message chunks

Resuming a Workflow Stream

Use getRun to reconnect to an in-progress or completed workflow:

typescript
import { getRun } from "workflow/api";

const run = await getRun(runId);
const readable = await run.getReadable({ startIndex });

Using Steps

Steps are durable checkpoints that persist their results:

typescript
async function getMessageHistory(chatId: string) {
  "use step";

  const dbMessages = await getChatMessages(chatId);
  return convertDbMessagesToUIMessages(dbMessages);
}

Streaming UIMessageChunks

When streaming UIMessageChunk responses to clients (e.g., chat messages), you must signal the start and end of the stream. This is required for proper stream framing with WorkflowChatTransport.

Always call startStream() before agent.run() and finishStream() after:

typescript
import { getWritable } from "workflow";
import { startStream, finishStream } from "../steps/stream";
import { chatAgent } from "@/lib/ai/chat-agent";

export async function chatWorkflow({ chatId, messageId }) {
  "use workflow";

  const history = await getMessageHistory(chatId);

  // Signal stream start with the message ID
  await startStream(messageId);

  // Run agent - streams UIMessageChunks to the client
  const { parts } = await chatAgent.run(history, {
    maxSteps: 10,
    writable: getWritable(),
  });

  await persistMessageParts({ chatId, messageId, parts });

  // Signal stream end and close the writable
  await finishStream();
}

The stream step functions write UIMessageChunk messages:

  • startStream(messageId) - Writes { type: "start", messageId } to signal a new message
  • finishStream() - Writes { type: "finish", finishReason: "stop" } and closes the stream

Without these signals, the client's WorkflowChatTransport cannot properly parse the streamed response.

Getting Workflow Metadata

Access the current run's metadata:

typescript
import { getWorkflowMetadata } from "workflow";

export async function chatWorkflow({ chatId }) {
  "use workflow";

  const { workflowRunId } = getWorkflowMetadata();

  // Store runId for resumption
  await createAssistantMessage({ chatId, runId: workflowRunId });
}

Workflow-Safe Logging

The workflow runtime doesn't support Node.js modules. Wrap logger calls in steps:

ts
import { logger } from "@/lib/logging/logger";

export async function log(
  level: "info" | "warn" | "error" | "debug",
  message: string,
  data?: Record<string, unknown>,
): Promise<void> {
  "use step";

  if (data) {
    logger[level](data, message);
  } else {
    logger[level](message);
  }
}

Running Agents in Workflows

Use the custom Agent class for full streaming control:

typescript
import { getWritable } from "workflow";
import { startStream, finishStream } from "../steps/stream";
import { chatAgent } from "@/lib/ai/chat-agent";

export async function chatWorkflow({ chatId, userMessage }) {
  "use workflow";

  const messageId = await createAssistantMessage({ chatId, runId });
  const history = await getMessageHistory(chatId);

  await startStream(messageId);

  const { parts } = await chatAgent.run(history, {
    maxSteps: 10,
    writable: getWritable(),
  });

  await persistMessageParts({ chatId, messageId, parts });
  await finishStream();
}

Persisting Workflow Results

Save agent output using step functions. The assertChatAgentParts function validates that generic UIMessage["parts"] (returned by agents) match your application's specific tool and data types:

ts
import type { UIMessage } from "ai";
import { insertMessageParts } from "@/lib/chat/queries";
import { assertChatAgentParts, type ChatAgentUIMessage } from "../types";

export async function persistMessageParts({
  chatId,
  messageId,
  parts,
}: {
  chatId: string;
  messageId: string;
  parts: UIMessage["parts"];
}): Promise<void> {
  "use step";

  assertChatAgentParts(parts);

  await insertMessageParts(chatId, messageId, parts);

  // Update chat timestamp
  await db
    .update(chats)
    .set({ updatedAt: new Date() })
    .where(eq(chats.id, chatId));
}

References