Developers
Drive your agents from your code, your terminal or your coding agent
An API described in OpenAPI, an MCP server for any coding agent, a CLI, and agents as open packages. Apache 2.0, installed on your servers in one command.
Coding agents
Any coding agent drives Appstrate
Claude Code, Codex, Cursor or another one: as soon as a coding agent speaks MCP, calls an API or runs a command, it can do everything. It runs an agent, waits for the result, reads the files it produced or imports a package, with your permissions and nothing more.
- Through your organization's MCP server, the REST API or the CLI: three doors to the same operations.
- Sign in through the browser (OAuth), or use an apst_ API key on a machine without one.
- An operation your role cannot perform is not listed, and every call is checked again when it runs.
- Every agent becomes a run-<agent> command, and published skills come through the MCP Skills extension.
- The CLI syncs your skills and agents into Claude Code and Codex (appstrate code sync).
Install
One command, then the tier that fits
The script downloads the CLI and checks its signature before putting it in place. appstrate install then asks for the infrastructure tier, generates the secrets and starts the stack.
- Tier 0: Bun and PGlite, no Docker and no isolation. For trying things out and development.
- Tiers 1 to 3: Docker Compose with PostgreSQL, then Redis, then S3-compatible storage.
- Tier 2 is the recommended one for a server. Plan for 4 GB of memory at least.
- macOS and Linux (x64 and arm64), Windows through WSL2. The same script runs unattended in CI.
API
A POST runs an agent, a GET waits for the result
Every instance describes its API in OpenAPI 3.1. An apst_ key is pinned to one space: no organization or space header to send, and it never holds more rights than the person who created it.
- The instance serves the description (/api/openapi.json), with a Swagger console at /api/docs.
- No official SDK: generate a client from the description, or call the API with appstrate api.
- An Idempotency-Key header makes a launch safe to retry.
- Embedding Appstrate in your product: the Appstrate-User header runs the agent for one of your users, with their own connections.
AFPS format
An agent is a folder you version in Git
A manifest.json declares the identity, the input and output schemas and the dependencies; a prompt.md holds the instructions. Skills follow the Agent Skills format: a folder with a SKILL.md, which also works in Claude Code.
- Published versions are immutable and follow semver. Dependencies are declared as ranges (^1.0.0).
- appstrate packages pull writes a package to a local folder, push sends it back to the draft, publish cuts the version.
- An agent exports with its skills and integrations as an .afps-bundle that another instance imports.
- The AFPS specification is published separately, under CC-BY-4.0.
Your system knows when a run is done
A run goes through known statuses. At each step that matters, a signed webhook goes to your URL (HMAC-SHA256, Standard Webhooks format), with retries and a delivery history.
What runs on your servers
One API process, one Docker sandbox per run, and the data services of your tier. Credentials stay in the sidecar: the agent never sees them.
Why Appstrate is open source
Transparency and trust
You can read every line that touches your data and your credentials.
Sovereignty
Appstrate runs anywhere, from a laptop to an air-gapped network, with no required account, no usage telemetry and no usage caps.
No lock-in
The same code runs on your servers and on Appstrate Cloud, and your agents export to the open AFPS format.
Our commitments
- A feature released under Apache 2.0 will never become paid.
- The core stays Apache 2.0, and so does every contribution: no relicensing.
- A feature planned for both editions ships in both on the same day.
- If a feature lands behind the license by mistake, we move it back to the core.
- If Appstrate stops maintaining the enterprise edition, it becomes open source.
Frequently asked questions
Is there an SDK?
Can an agent read my API keys or OAuth tokens?
Can an incoming webhook trigger an agent?
Can I run an agent from CI?
Which tier for production, and what should I plan for?
Can I leave with my agents?
How stable is it?
Let's see what it would look like on your side.
Try it on your own in a few minutes, or take thirty minutes with us to scope a first use case.
Let's talk about your case
Thirty minutes with us to scope a first use case and its gain.
Read the code
Apache 2.0. Your technical team can read it all, install it all, keep it all.