Watch an agent become a command.
No mockups, no edits. A real terminal, a live account: a2a agents lists the registry, a2a use turns an agent into typed shell commands, and a real call wakes a scale-to-zero agent that answers on camera.
- →a2a agents — the live registry: email, DNS, payments, scrapers, reviewers
- →a2a use my-first-agent — tools become typed commands with flags
- →a real call: the agent wakes, runs on an LLM, answers in your shell
From empty box to deployed agent
Every video starts at a real login prompt in a clean Ubuntu container. What you see is the product, current as of the latest release.
Zero to agent project in one command
From a clean Ubuntu login to a complete agent project: a2a init scaffolds the manifest, agent code, and requirements — ready for a2a dev, a2a test, and a2a deploy.
- →real login on a clean box — no prebaked environment
- →a2a init research-agent scaffolds manifest + agent.py + requirements
- →next steps printed: dev, test, deploy
The agent-calling vocabulary
The four commands that make agents feel native: a2a use for typed stubs, a2a call as the schema-free escape hatch, a2a chat for a terminal conversation, a2a deploy to ship.
- →a2a use — typed commands from any agent's card
- →a2a call agent tool key=value — no setup, no schema
- →a2a deploy — ship it when it works
“Demo videos age. Ours re-render from scripts against every release — if it plays, it ships.”
What you're actually watching
Are these demos real or simulated?
Real. Each video is an unedited screen recording of the a2a CLI running in a clean Ubuntu container — a real login prompt, real commands, and in the marketplace demo a live agent call against a production account. The narration script and the recording harness are versioned in the platform monorepo, and the videos are re-rendered against the current product, so what you watch is what ships.
What is 'agents as commands'?
Run `a2a use <agent>` and the CLI fetches the agent's card — its tools and input schemas — and caches a typed stub. From then on the agent's tools are native shell commands with flags: `a2a my-first-agent ask --prompt ...`. For one-off calls without a stub, `a2a call <agent> <tool> key=value` is the schema-free escape hatch.
What happens when I call an agent that's scaled to zero?
It wakes. Agents on a2a cloud scale to zero between calls; a call cold-starts the service, streams progress to your terminal while it runs, and returns the result. You only pay for what runs, and every run is Ed25519-signed so you have a receipt for what happened.
How do I get from watching this to my own agent?
Three commands: `pip install a2a-pack`, `a2a init my-agent`, `a2a deploy`. The scaffold includes the manifest, a starter agent with typed tools, and an AGENTS.md so your coding agent already knows the SDK. The quickstart takes you through it end to end.
Your agent could be the next recording.
pip install a2a-pack, a2a init, a2a deploy — a managed Postgres database, an MCP server, an API, and an Ed25519-signed receipt for every run. From empty terminal to live agent in the time it took to watch the hero video.