AI agents write code. Meatloaf lets them build, run, test, and ship it. One CLI to bridge the gap between generating code and delivering working software.
Language models generate thousands of lines of code but have no way to execute, test, or see the results. They're all brain, no body.
No way to spin up a server, connect to a database, or interact with running processes.
No visual feedback. Can't take screenshots, check renders, or verify what the app looks like.
No ability to hit endpoints, check responses, or verify that code actually works.
No real-time feedback loop. Can't read logs, catch errors, or iterate on failures.
Meatloaf gives AI agents full control over the build-test-ship lifecycle through isolated Docker sandboxes.
Isolated Docker containers. Create, destroy, start, stop.
Run commands, scripts, upload and download files.
Hit endpoints, batch test, load test, wait for readiness.
Capture what the app actually looks like. HTML snapshots too.
Stream logs, resource stats, live terminal dashboard.
Encrypted secrets management. Inject into sandboxes.
Docker networks, port forwarding, tunneling.
Persistent storage that survives container restarts.
Clone, commit, push — full git inside sandboxes.
Scheduled tasks running inside your sandboxes.
Multi-container orchestration, Docker Compose style.
JSON output, auto-detect, works with any AI agent.
| Command | Description |
|---|---|
meatloaf create | Create a new sandbox |
meatloaf exec <id> <cmd> | Run a command in a sandbox |
meatloaf test --url <url> | HTTP request tester |
meatloaf screenshot <url> | Capture a screenshot |
meatloaf up ./project | Full lifecycle: build, run, test |
meatloaf batch <file> | Run batch tests from JSON |
meatloaf logs <id> | Stream container logs |
meatloaf shell <id> | Interactive shell |
meatloaf compose up | Multi-container orchestration |
meatloaf secrets set | Manage encrypted secrets |
meatloaf snapshot save | Save and restore state |
meatloaf bench <url> | Load testing with percentiles |
npm install -g meatloaf-cli