Build your first agent

Your first agent

An agent needs three things: a name, a model, and a prompt. That's a complete, runnable agent.

agent: greeter
model: claude-opus-4-7
prompt: |
  Greet the user warmly in one sentence.

toac build greeter.agent emits a typed greeter.ts exporting a runnable greeter.

Typed inputs

Declare parameters under inputs, then reference them in the prompt with {inputs.<name>}. The count in inputs[1] must match the rows.

agent: greeter
model: claude-opus-4-7
inputs[1]{name,type}:
  name,string
prompt: |
  Greet {inputs.name} warmly in one sentence.

The generated agent takes a typed { name: string } — pass the wrong shape and it's a compile error.

Structured output

Add outputs and the agent returns a typed, validated object instead of free text. Arrays use [].

agent: extractor
model: claude-opus-4-7
inputs[1]{name,type}:
  text,string
prompt: |
  Extract the key facts from: {inputs.text}
outputs[2]{name,type}:
  title,string
  facts,string[]

toac emits a zod schema and a "respond" tool, so the result is parsed and typed as { title: string; facts: string[] }.

Tools

Name tools under tools; implement them in a co-located <agent>.tools.ts. The runtime runs the tool-use loop for you.

agent: researcher
model: claude-opus-4-7
inputs[1]{name,type}:
  topic,string
tools[2]: web_search,fetch_page
prompt: |
  Research {inputs.topic}. Use web_search, then fetch_page to read sources.
outputs[1]{name,type}:
  summary,string

// researcher.tools.ts
import { defineTool } from "toad-runtime";
import { z } from "zod";

export const web_search = defineTool({
  description: "Search the web for a query",
  input: z.object({ query: z.string() }),
  run: async ({ query }) => `results for ${query}`,
});

Loops & conditionals

The prompt is a small, type-checked template: {#each} over array inputs and {#if} on booleans.

agent: report
model: claude-opus-4-7
inputs[2]{name,type}:
  findings,string[]
  detailed,boolean
prompt: |
  Write a report from these findings:
  {#each inputs.findings as f, i}
  {i}. {f}
  {/each}
  {#if inputs.detailed}
  Add a thorough analysis section.
  {:else}
  Keep it to one paragraph.
  {/if}

Loops compile to .map().join(""); conditionals to nested ternaries — all validated against the typed inputs.

Composition

Declare another agent under uses and toac wires it in as a tool — one agent calls another.

agent: planner
model: claude-opus-4-7
uses[1]: researcher
prompt: |
  Plan an article. Use the researcher tool to gather sources first.

The sub-agent's typed inputs become the tool's input schema automatically. That's the whole framework: small files that compose.