Quickstart

This is the local happy path. For server deployments, Docker Compose, or driving an instance from another machine, see Install.

Install an instance

curl -fsSL https://get.appstrate.dev | bash

The script installs the appstrate CLI. In an interactive terminal it then asks you to run the installer yourself:

appstrate install

Pick a tier when asked. Tier 2 (PostgreSQL + Redis, agents in Docker containers) is the default when Docker is running. Tier 0 needs nothing but Bun: an embedded database, and agents as plain subprocesses, which is fine for evaluation. When the healthcheck passes, the instance is at http://localhost:3000.

To skip the prompts, add --yes: curl -fsSL https://get.appstrate.dev | bash -s -- --yes. That also makes the instance invitation-only on Docker tiers, so you claim ownership at /claim with the token the installer prints. See Install.

Create your account, organization and model

Open http://localhost:3000 and sign up (or claim the instance). The onboarding flow guides you through:

Create your organization. A default space and a hello-world demo agent are created with it.
Add an AI model. Agents and the chat need at least one. Choose a provider (Anthropic, OpenAI, OpenRouter, Mistral, Groq and others, or any OpenAI-compatible or Anthropic-compatible endpoint), paste an API key, pick one or more of the models it lists and save.
Invite users (optional).

You can skip the model step and add one later under Organization settings → Models.

Run your first agent

The hello-world agent starts as a draft, with no published version. The webapp's Run button and the CLI's @draft run that draft, because you can edit the agent. The chat and a coding agent run the latest published version by default, and the /appstrate:run-hello-world command only appears once one exists: for those, open the agent first and choose Create version in its actions menu.

Open Agents and select Hello World.
Click Run.

The run streams live: logs, tool calls, and the result when it finishes. Open Runs to see the history with status, duration and cost.

Open Chat in the sidebar, pick a model, and ask for what you want, for example "Run my Hello World agent and show me the result". The assistant acts with your own permissions, launches the run and reports back. See Chat.

appstrate login --instance http://localhost:3000   # approve the code in your browser
appstrate api GET /api/agents                      # each agent has an id like @acme/hello-world
appstrate run @acme/hello-world@draft              # launch the draft and follow the logs

appstrate login pins your organization and its default space on the profile, so later commands need no extra flags. Replace @acme with your organization's slug. See CLI.

With Claude Code, install the Appstrate plugin once. Every agent active in your space then appears as a /appstrate:run-<agent> command, and your space's skills as /appstrate:<skill>:

appstrate login --instance http://localhost:3000
claude plugin marketplace add appstrate/claude-plugins
claude plugin install appstrate@appstrate

Then, in a new Claude Code session, run /mcp and authenticate the plugin:appstrate:appstrate server, and try /appstrate:run-hello-world. See Skills and coding agents.

On Docker tiers the agent ran in its own container, and the credentials of any integration it used stayed in the sidecar next to it, never inside the agent process.

What you have now

A local Appstrate instance on port 3000, with your organization and default space
A model connected to the organization
One agent you have run, and a way to run more: the webapp, the chat, the CLI, or a coding agent

Where to go next

Concepts: the mental model behind agents, runs, integrations, spaces and packages
Using Appstrate / Webapp: a tour of the dashboard
Using Appstrate / CLI: the command reference
Features: one page per primitive
Self-Hosting: when to leave Tier 0 and how to operate in production