Description
Architecture
Skills with capsule integrated
Knowledge graph
Handover
capsule integrated chat

CAPSULE

Inspiration

Every AI coding session creates valuable knowledge, but that knowledge disappears when the session ends. We built CAPSULE to preserve those lessons so every developer and AI agent starts smarter instead of starting from scratch.

What it does

CAPSULE captures AI coding sessions, distills reusable knowledge locally using Ollama, scores and versions it, and stores it as trusted enterprise skills that can be reused across the organization.

How we built it

Built with Next.js, React, TypeScript, Node.js, Ollama (Qwen2.5-Coder 14B), Backboard, Docker, and Git. We created an end-to-end pipeline for capture → distill → score → version → promote.

Challenges we ran into

Separating valuable knowledge from conversation noise.
Building reliable evaluation and governance.
Keeping all sensitive code local and private.

Accomplishments that we're proud of

✅ Built a working end-to-end prototype.
✅ Captured real AI coding sessions into reusable skills.
✅ Demonstrated measurable productivity gains through A/B evaluation.

What we learned

Enterprise AI needs more than memory—it needs trust, provenance, and validation. Local LLMs are powerful enough to enable private, low-cost knowledge capture.

What's next for CAPSULE

We plan to integrate CAPSULE directly into AI coding assistants like Claude Code, Cursor, and GitHub Copilot, expand enterprise collaboration features, strengthen automated evaluation, and become the persistent organizational memory layer for AI-assisted software development.