You build the agent.
We run the loop.

Ship agents inside your product. Cereal handles the model calls, tools, and durable sessions—so you can focus on what your agent does.

Explore a loop See the API

Your models. Your keys.Bring the providers you already use.

Your tools stay with you.Connect your own tools over MCP.

Every turn, in the open.Follow the session from start to finish.

A little definition.
A whole lot handled.

Define your agent. Start a session. Follow every turn.

{
  "name": "support-agent",
  "system": "Help customers with their orders.",
  "model": {
    "provider": "your-provider",
    "model": "your-model"
  },
  "tools": [{
    "type": "mcp",
    "name": "orders",
    "url": "https://your-app.com/mcp"
  }]
}

An agent loop, made visible

Message received“Where is my order?”
Model respondsDecides which tool to call
Tool calledLooks up the order in your app
Result returnedUses the result to answer
Ready for the next turnSession history stays together

Illustrative session · no live model calls

Your tools run on your infrastructure. Cereal manages the session.

Versioned definitions · HTTP API · MCP tools · Event streams

Example values are yours to configure. Follow the quickstart

Keep your stack.

Connect your models and MCP tools. Call your agents over HTTP from the product you already have.

Stay in control.

Pin agent versions, review tool approvals, and set spend limits before a session runs.

See the whole loop.

Inspect model calls, tool results, and usage in one session trace.

Good agents deserve
a better runtime.

Open the console

Bring your model key. Make your first agent.

You build the agent.We run the loop.

A little definition.A whole lot handled.