Evaluate agent work with evidence.

Prepare separate task directories, import externally produced results, and grade them against reviewer-defined criteria. Compare runs under matched conditions.

Evaris example output
$ go run ./examples/offline

good.txt: ordinary tests PASS,
          independent assessment PASS

bad.txt:  ordinary tests PASS,
          independent assessment FAIL

Offline grader controls. Both ordinary suites pass.

Quick start

Install the CLI

Use npm with Node.js 18+ on Linux or macOS, or download an archive from GitHub.

Check the npm release
npx --yes evaris-cli@latest --version

Install for a project or globally · Archives

Run the offline demo

Use Go 1.27.4+ on your PATH. The demo grades authored controls and prints the report path.

Run the offline demo
git clone https://github.com/refactorroom/evaris.git
cd evaris
go run ./examples/offline

The demo runs locally without a model API.

Choose a workflow

Reference

Keep reviewer inputs private and run agents in separate task directories. Use --allow-commands only for trusted graders and verifiers. The Claude Code transcript importer and explicit trial pairing are documented on the development branch.