Calypso AI

Inspiration

Creating product tutorials is slow and repetitive. Teams must record workflows, remove delays, add narration, and redo the entire video whenever one step changes. Calypso AI automates this process and makes tutorials easy to update.

What it does

Calypso turns a browser task into a polished tutorial video.

It explores the website, verifies the correct workflow, records a deterministic replay, divides it into scenes, generates narration, synchronizes audio with clicks and typing, removes idle time, and produces the final video.

Users can also update a single scene without rebuilding everything. Narration-only edits reuse the existing recording, while action changes regenerate the affected scenes.

How we built it

Calypso uses Browser Use Cloud and Playwright for verified browser automation and recording.

OpenAI models handle browser reasoning, workflow verification, scene planning, and narration writing.

GenBlaze runs timestamped ElevenLabs narration jobs for each scene in parallel and generates thumbnails.

Backblaze B2 stores narration, recordings, final videos, and immutable revision artifacts. FastAPI, PostgreSQL, Redis, and Celery power the production API and job system, while FFmpeg handles video composition.

Challenges we ran into

The main challenge was removing AI thinking time from recordings. We solved this with two passes: one for exploration and one deterministic pass for recording.

We also had to handle unstable webpage elements, synchronize narration with browser actions, and safely update scenes without breaking later steps.

Accomplishments that we're proud of

We built a system that creates tutorials from real, verified browser interactions.

Calypso supports deterministic replay, scene-level editing, timestamped narration, secure B2 storage, and immutable revisions that keep the current video live until an update is validated.

What we learned

We learned that reliable browser automation requires reproducible actions, strong verification, and structured scene data.

We also learned that tutorials are easier to maintain when they are treated as editable productions rather than static recordings.

What's next for Calypso AI

Next, we plan to add a visual scene editor, multilingual dubbing, subtitles, custom branding, team collaboration, faster partial rendering, and automatic updates when a website changes.

Built With

• backblaze
• digitalocean
• fast-api
• genblaze
• vercel