01 Open-source AI testing agents
Agents that test your software. Not agents you have to test.
An open-source AI testing agent uses AI to plan, write, run, or repair tests for your application. That is a different job from an agent-evaluation tool that scores how well a separate AI agent behaves.
If you are looking to grade an LLM agent's own responses, tools like agentevals or LangWatch cover that different intent. OpenEvident's agents test your application; they do not judge another agent.
OpenEvident's agents follow one loop: ground a real flow in the browser, generate a Playwright automation agent script from that grounded evidence, prove the result against your own service logs, and heal the suite as the product changes. Vindicate runs this loop locally over MCP, so Cursor, Claude Code, and GitHub Copilot can call it directly — no dashboard, no queued job on someone else's infrastructure.
Most AI-native testing platforms are SaaS: your test data, your app's flows, and your evidence all live in someone else's cloud. OpenEvident's agents are local-first AI testing agents, released under Apache 2.0 — the code runs on 127.0.0.1, and nothing leaves your machine unless you wire up your own CI runner.
01
Where it runs
On your machine over a local MCP server — not a hosted job queue you have to trust with your app.
02
Who owns the evidence
You do. Recordings, generated specs, and logs stay in your repo and your account.
03
How it plugs in
Directly into the coding agent you already use — Cursor, Claude Code, or GitHub Copilot — over MCP.
04
License
Apache 2.0 on every public repository. No feature gated behind a login.