Skip to main content
anpord is the TypeScript package for defining and running coding-agent evals. Cases, mocks and validators live in your repository alongside the code they test.
Start with the quickstart to define a suite and run it on Anpord.

​
Package exports

Definitions take objects. Handlers use ordinary TypeScript functions and promises. Standard Schema infers mock inputs and checks values at runtime; examples use Zod. You do not need Effect imports or a separate model client.

​
One definition, multiple ways to run

Keep suites in *.eval.ts files. The CLI compiles them locally, including imported fixtures and validators, then submits them to Anpord’s hosted infrastructure:
The GitHub Action runs the same command. To submit from application code, use compileEval with the API client. Results retain the calls, validator evidence and captured source for that run.

​
Connections

Configure ANPORD_API_KEY for calls to Anpord. Save model credentials in Settings > Harnesses; they do not belong in eval definitions. Hosted sandboxes and mocked product services need no separate keys by default. For agent setup, start with Claude Code or the supported agents.