Build a daily brief agent, step by step

A daily brief is the “hello world” of useful agents: every morning, a short summary of what needs attention, drawn from the user’s own messages and email. It touches every part of Agent200, so it is a good first build. Here is the whole path.

What you are building

Each morning, for each user who opted in, your app asks Agent200 for a brief. Agent200 gathers the relevant messages and email for that user, a model writes a short summary, and your app delivers it wherever your users read it.

Step 1: Set up the project

Create an Agent200 project, generate an API key for your development environment, and enable the chat and email services your users rely on, with read-only endpoints.

Step 2: Create the brief workflow

Configure one capability that:

1. retrieves recent relevant chat messages,
2. retrieves recent relevant email,
3. returns both in one structured result.

Now the model has one tool, brief, instead of several endpoints to coordinate.

Step 3: Connect each user’s accounts

The first time a user turns the brief on, Agent200 needs their authorization. It returns an authorization URL, or sends them through hosted authorization, and stores the connection for that user. From then on, every brief for them runs on their own accounts.

Step 4: Make the request

Conceptual example.

const brief = await agent200
  .user(user.id)
  .run({
    model: "openai/gpt-5.6-sol",
    instructions: `
      Write a morning brief in five bullet points or fewer.
      Lead with anything that needs a reply today.
      Today is ${today}.
    `,
    input: "Prepare my brief.",
    tools: ["brief"],
  });

Step 5: Schedule it in your app

Agent200 runs when you call it. Your own scheduler, a cron job or a queue worker, starts the request each morning in the user’s time zone. That keeps timing, retries and opt-outs in your hands.

Make it good

• Give the date. “Today” and “this week” only work if the model knows when today is.
• Set a shape. A fixed number of bullets is easier to read every morning than free prose.
• Lead with action. Ask for what needs a reply first; summaries second.
• Let users steer. Store a sentence of preferences per user and add it to the instructions.

Where to take it next

Add the calendar for the day’s meetings, or iGPT for answers from long threads and attachments. Each addition is another service in the workflow, not another integration.

See more ideas on the Use Cases page.