Kane CLI - Natural language browser & mobile app tests from terminal
by•
An agentic quality verifier for developers and AI coding agents. Describe a test in natural language, and Kane CLI runs it in a real Chrome browser and returns pass or fail with shareable proof. No selectors to write. Local-first, free to start.
Replies
Launch day energy is real today 🎉
I have told the Kane story on stages and in campaigns for months, and the reaction is the same every time. People assume they will need to write code or learn a framework. Then they watch someone type a sentence in natural language, a real browser opens, and it comes back with an evidence pack and a verdict.
That moment when the room goes "wait, that's it?" never gets old.
And today the story got a new chapter: the same commands now run on Android emulators and iOS simulators too.
Even outside engineering, I use Kane CLI for my own repetitive browser work. That is the part I love most, it is not just for engineers.
The Starter plan is free, so go break it and tell us what you think. The team is here in the comments all day.
Learn more: https://www.testmuai.com/kane-cli-ph/
Awesome, i have been needing to formalise our test releases.. this is just what i was looking for.
@lachlan_greenbank Love that Lachlan 🙏 That's the exact gap we built for. Tests live as _test.md in your repo, every run comes back with an evidence pack, and you can gate the deploy on it.
@lachlan_greenbank Give it a go and let us know your feedback :)
Does Kane handle logged-in flows automatically, or do we need to set up auth ourselves?
@deepak144 Both work. You can log in as part of the objective itself, just describe it in natural language with your credentials as variables. Or for flows that depend on an existing session, use a named Chrome profile: run kane-cli config chrome-profile, log in once, and every run after that starts already authenticated. No auth code to write either way.
Agentic mobile test was the missing piece of the puzzle. Kane CLI seems to be solving that perfectly well!
@misbah_farooqi Thanks a lot Misbah 🙏🚀
the machine-readable verdict for agent loops is the right instinct, that's the same gap I hit running voice model evals: a pass/fail with no provenance just means trusting the model again. does the evidence pack capture intermediate state, or just the final screenshot, when a flaky animation causes a false fail?
@sabber_ahamed Great instinct, and it goes well beyond a final screenshot. The evidence pack captures per-step screenshots (with annotated copies), a full HAR network log, console output as NDJSON, and failure records, all replayable step by step. So on a flaky animation you can see exactly which intermediate state the check ran against, not just the end frame. Provenance is the whole point: a verdict you cannot trace is just trust with extra steps.
AI coding agents desperately need something like this. Writing code is becoming easy; reliably verifying the result is becoming the bottleneck.
@subhash_kanagamani Exactly the thesis we built on. The bottleneck moved from writing code to trusting it. That is why Kane CLI is agent-native out of the box: Claude Code or Cursor can call it, read the verdict from structured output, and close their own loop. Would love to hear how you are handling verification today.
Cool! We use our own skills for this in Claude Code.