miroir
Health Pass
- License — License: Apache-2.0
- Description — Repository has a description
- Active repo — Last push 0 days ago
- Community trust — 39 GitHub stars
Code Pass
- Code scan — Scanned 12 files during light audit, no dangerous patterns found
Permissions Pass
- Permissions — No dangerous permissions requested
No AI report is available for this listing yet.
Turn a product into an agentic RL environment
Miroir 🪞
Turn a product into an agentic RL environment.
Miroir packages real applications into environments for training and evaluating agents.
Add a manifest and tasks to your product's Docker image. Miroir handles the rest: seeding data, resetting state between episodes, exposing APIs as tools through OpenEnv and MCP, and grading what the agent actually changed—including database rows, emails, and HTTP requests.
Quick start
Requirements: Docker with buildx and uv.
1. Build and verify
docker buildx bake acme-crm-env
docker run --rm --network none acme-crm-env check
This builds Acme CRM, a sample application with Postgres, Redis, a web server, a worker, and an MCP server. The check command replays every task's reference run and verifies its expected outcome.
2. Serve the environment
docker run --rm -d -p 127.0.0.1:8000:8000 \
-e MIROIR_TOKEN=secret acme-crm-env
curl http://127.0.0.1:8000/health
3. Run an episode
Install the Python client:
uv venv && source .venv/bin/activate
uv pip install miroir
Reset an episode, execute an action, and inspect its outcome:
from miroir import Session
session = Session.lease(
["http://127.0.0.1:8000"],
token="secret",
wait_busy=False,
)
start = session.reset("pay-invoice")
print(start["instruction"])
step = session.step(
"api_request",
{"method": "POST", "path": "/invoices/4001/pay"},
)
print(step["observation"]["result"]["status"])
outcome = session.finish()
print(outcome["reward"], outcome["metadata"]["checks"])
session.close()
The episode receives a reward of 1 only if the invoice is paid.
Documentation
- Introduction
- How Miroir works
- Build an environment
- Train and evaluate
- Run in Harbor: Claude Code, Codex and other agents on an env's tasks
Gitea example — Unmodified Gitea with 144 tasks across 12 task families.
Build the documentation locally with mdbook serve docs.
Coding agents
Use the Miroir skill to build environments, write tasks, debug checks, and run training with coding agents.
Claude Code
/plugin install miroir --marketplace raphaelDkhn/miroir
Other agents
npx skills add raphaelDkhn/miroir
Contributing
See CONTRIBUTING.md. Report security issues privately, as
SECURITY.md describes.
License
Reviews (0)
Sign in to leave a review.
Leave a reviewNo results found