Every product needs a walkthrough video, and making one takes hours. You write a script, record the screen, retake it, record the voice-over and sync the captions. Then you ship one update and the video is already out of date.

Walkcast does it from a single URL. It opens a real Chromium browser, signs in if you give it credentials, walks through your pages and records what actually happens. Every line of narration is written from recorded events, not from a script and not guessed by an AI. You get:

• recording.mp4 with a spoken voice-over (H.264 + AAC)

• captions.vtt timed to every line

• A clickable transcript with per-page chapters

• narration.wav, plus a full action trace

You pick the voice: the free built-in one, ElevenLabs, OpenAI, Google, Azure, Polly, Sarvam, and more, or your own microphone. Each voice can be sampled before you record.

Walkcast Lab uses the same recorder for a second job: debugging browser AI agents. It turns a failed agent run into a reproducible regression test. The cause is classified from the application’s real state, not from the agent’s claim or an LLM judge, and anything the rules can’t explain is marked “unexplained”. Replay that test on the next agent version to see whether it fixed the bug or broke something else.

Built-in guardrails:

• You must confirm you own the site or are authorized to test it.

• It never bypasses CAPTCHA or 2FA. It stops and says so.

• It honours robots.txt.

• Passwords are never written to disk or spoken in the narration.

Open source under the MIT license, and it costs nothing. Available for Windows, macOS and Linux.

Categories / tags
Developer Tools, Productivity, Open Source, AI, Video, Testing, Marketing

Links

• Website: https://walkcast.kunjshah.in

• GitHub: https://github.com/KunjShah95/walkcast