Workflows: configure a capability once, call it by name

Agents rarely need a single endpoint. A morning brief needs recent messages and recent email. Meeting research needs a customer record, the related email and the conversations around it. Workflows let you put those operations together once, give the result a name, and call it from anywhere in your application.

From endpoints to a capability

A workflow combines endpoints from one service or from several into a single named capability. You configure it in the Agent200 dashboard or through the SDK. From then on, the model sees one tool with a clear name, a description of what it does and the inputs it accepts. It does not see, or need to understand, the HTTP calls behind it.

That separation keeps tool definitions small and readable for the model, and keeps the integration details in one place for you.

Example: brief

A capability named brief gathers what someone needs to start their day:

1. Retrieve the relevant chat messages.
2. Retrieve the relevant email.
3. Combine both into the output structure you defined.
4. Return the result to the model.

To use it, reference it by name.

Conceptual example.

tools: ["brief"]

What comes back

A workflow returns structured data the model can work with directly. The exact shape follows the endpoints and the output you configure:

{
  "messages": [
    { "message": "Project update..." }
  ],
  "email": [
    { "subject": "Weekly update", "content": "..." }
  ]
}

Aggregate or chain

Aggregated

Several endpoints run and their results are combined. Messages and email are fetched independently and returned as one result.

Sequential

The output of one step feeds the next. A workflow can look something up in one service, pull out an identifier, and pass it to another service.

An illustrative configuration: customer_context

1. Retrieve the customer’s details from the CRM.
2. Retrieve related email.
3. Search the relevant team conversations.
4. Return the combined context.

Before a sales call, one tool call gives the model everything above, for the user the request runs as.

Why naming matters

A named capability is reusable. The same brief can serve a daily summary, an on-demand question and a weekly report, and when you improve it, every caller gets the improvement. That is the difference between building an integration and building a library of capabilities your agents share.

See how workflows work in more detail.