RedCrown documentation
Check a change to an AI feature using your own cases, requirements, and available evidence. RedCrown scores the outputs your feature produced, or runs an evaluation for you on your keys, applies the requirements you wrote down, and keeps one record per run with the cases, the counts, the exclusions and the caveats. These docs cover the redcrown CLI, the app, the MCP tools, and the harness commands that grade a deployed AI feature case by case.
Start with what you want to do
| I want to | Go to |
|---|---|
| Try a regression example without provider keys | Check your first AI change, the quickstart. Offline, no account. |
| I already have outputs | Score outputs produced elsewhere, section 2 of the baselines guide: supplied-output scoring, and which form can be checked against a reference run. |
| I already have a test harness | Start from what you already have: a journal, production traces, or a rules list, reported and compared offline. |
| I want to accept a baseline in the app | Accept a baseline, section 3 of the baselines guide: who accepts, what is stored, what is refused. |
| I need to compare models | Compare models on your workload: run every candidate on your cases and rank them. |
| I need production capture or replay | Live proxy and collect: a separate advanced workflow, Growth and up. |
| I want the check in CI | Run the check in CI: the boundary, the exit codes, a shell gate and a GitHub Actions job. |
Start here
- Check your first AI change: the synthetic regression example, end to end, offline.
- Concepts: AI feature, requirements and cases, reference run, accepted baseline, candidate, check result, human review, retained evidence.
Guides
- Accepted baselines and regression checks: score supplied outputs, check a later version offline, accept a baseline in the app.
- Run the check in CI: preserve the exit status and let your pipeline decide.
- Start from what you already have: a journal, production traces, or a rules list, graded offline.
- Harness guide for coding agents: exact commands, exit codes, and what never to claim.
- Compare models on your workload
- Try a sample
- Run a Face-Off
- Bring your own data
- Connect providers and credentials
- Run and rank an experiment
- Inspect runs and reports
- Import external eval results
- Share a proof
- Signed attestation
- Reviewer portal
- Live proxy and collect mode
- Cost simulator
- Benchmarks
- Workspaces and billing
- Authentication
Reference
For agents
An LLM can read /llms.txt (the index) or /llms-full.txt (the whole corpus), and any page has a Markdown twin at its index.md.
Last verified .