Get Started

Quickstart

From zero to your first agent run in 5 minutes.

Install and run

Three steps, local dev happy path. For server deploys, curl one-shot, Docker Compose, or driving an instance from another machine, see Install.

Install the CLI and spin up a local instance

bun install -g appstrate && appstrate install

Interactive tier picker. Opens http://localhost:3000 once the healthcheck passes.

No Bun, deploying on a server, or driving from another machine? See Install for curl one-shot, Docker Compose, and multi-machine setups.

Sign in

appstrate login

The CLI prints a short user-code + URL, opens the URL in your browser. You sign up (first user owns the org), approve the device, and the exchange writes a 15-minute access token + 30-day rotating refresh token to your OS keyring.

Connect a model and run your first agent

Install the Appstrate Agent Skill into Claude Code / Cursor / Antigravity or any AGENTS.md-aware host. One command auto-detects your target:

curl -fsSL https://raw.githubusercontent.com/appstrate/skills/main/install.sh | bash -s appstrate

Then delegate in plain English. Example prompts to get you from zero to a first run:

  • Connect a model:

    Connect my Anthropic API key so I can run agents against Claude.

  • Build and run:

    Build an agent called hello-world with prompt Summarize this URL: {url}, then run it against https://anthropic.com and show me the output.

  • Inspect:

    Show me the last 5 runs with their status and duration.

The skill teaches your coding agent the CLI, AFPS packaging, scoped names, profiles, and the REST API. Full reference: Using Appstrate / Skill.

Connect a model. You need at least one LLM key before running an agent.

  1. Open the webapp at http://localhost:3000
  2. Go to Settings → Models, click Add a model
  3. Pick a provider (OpenAI, Anthropic, OpenRouter, any OpenAI-compatible)
  4. Paste your key, save

Run the built-in hello-world agent. A hello-world agent is provisioned on a fresh install.

  1. Open the hello-world agent from the agents list
  2. Click Run, fill the input, hit go

You should see a streamed response within seconds. The container is created for the run and destroyed when it finishes. Credentials never reach the agent process: they pass through a sidecar proxy.

What you have now

  • A local Appstrate instance at :3000
  • An authenticated CLI on your machine
  • One working agent with one connected model
  • Either a webapp you drive by hand, or a coding agent that drives the CLI for you

Where to go next

Introduction

Everything you need to ship autonomous AI agents with Appstrate.

Install

Where to install Appstrate, how to drive it, and how to do both from two different machines.

On this page

Install and runInstall the CLI and spin up a local instanceSign inConnect a model and run your first agentWhat you have nowWhere to go next