GLAT
Shared memory for AI coding assistants, built from teammates’ unmerged code changes.
- Role
- Built end to end with two teammates
- Team
- 3 people
- When
- Mar – Apr 2026
- Outcome
- Public release across 3 versions
- Built with
- TypeScript, Supabase, Gemini, VS Code API
The problem
Your AI agent doesn’t know what your teammate’s agent is doing. You end up passing session files, writing context notes and explaining work by hand just to keep them in sync.
What we built
I designed and built GLAT end to end with two teammates, giving coding assistants a shared memory across branches. It connects a task to the team’s relevant work in progress and supplies the actual diffs before those changes are merged.
More detail
I designed the retrieval pipeline to find changes beyond the file being edited: Gemini predicts affected files, so an assistant editing a caller can discover a teammate’s upstream API change. Search matches summaries, then retrieves full diffs from Supabase. GitHub repository permissions control access.
The result
Released publicly as an installable VS Code extension across three versions.
We demonstrated coordination across separate frontend and backend branches: the frontend assistant built against its teammate’s new API response format before the backend change was merged.