Getting started

At the end of this page you've run an agent from code, both as a one-shot call and as a live stream.

Prerequisites — Node ≥ 22. A provider key: set one of ANTHROPIC_API_KEY / OPENAI_API_KEY / GOOGLE_API_KEY.

1. Install

bun add @ownware/loom      # bun
npm install @ownware/loom  # npm

@vscode/ripgrep auto-installs the correct platform binary for the grep tool via postinstall — no manual setup.

2. One-shot run

import { Loom } from '@ownware/loom'

const result = await Loom.run('anthropic:claude-sonnet-4-6', 'What is 2+2?')
console.log(result.text)   // "4"
console.log(result.usage)  // { inputTokens, outputTokens, cacheReadTokens, model, costUsd, ... }

npx tsx run.ts

3. Stream events instead

Loom's loop is an AsyncGenerator<LoomEvent> — filter in a for await:

import { Loom, filesystemTools } from '@ownware/loom'

for await (const event of Loom.stream('sonnet', 'Summarize package.json', {
  tools: filesystemTools,
})) {
  if (event.type === 'text.delta') process.stdout.write(event.text)
  if (event.type === 'tool.call.start') console.log(`→ ${event.toolName}(${JSON.stringify(event.input)})`)
}

'sonnet', 'opus', 'haiku' are short aliases — see Providers.

Verify it works

Run the streaming script; you should see the model's text print incrementally, and a → readFile(...) line when it reads package.json. That's the loop calling a tool and streaming the result.

Next steps

• Call patterns — one-shot, streaming, builder, and full sessions.
• Built-in tools — the filesystem/shell/browser kit.
• Streaming events — the full event vocabulary.