Three runnable scripts that show Computer, then Agent, then Crew. Each prints what it does and what came back.
You need a running Docker daemon (every example boots a real container) and OPENAI_API_KEY (the examples use createOpenAIProvider(), which defaults to gpt-4o). Without the key, each example prints SKIP and exits cleanly.
pnpm install && pnpm build # from the repo root
cd experimental/agents
export OPENAI_API_KEY=sk-...
node examples/single-agent.mjs
node examples/manager-crew.mjs
node examples/networked-crew.mjsRead them in order; each adds one idea.
single-agent.mjs:createAgent()boots aComputerfromapps/filesystemand returns anAgentwhose tools are that app's exports.manager-crew.mjs: oneComputerwithapps/filesystemandapps/notes, one workerAgentper app, and a manager that delegates to them withCrew.withManager()(each worker becomes a tool viaAgent.asTool()).networked-crew.mjs: the same delegation, but each worker runs its own agent loop in its ownComputer, booted withbootNetworkedAgent()on a shared Docker network and reached throughCrew.networked().
Full API: agents reference.