Project Intelligence
A learning system for Claude Code that gets smarter within every project over time. Archives sessions, generates searchable summaries, synthesizes reusable skills, tracks how your understanding evolves through reasoning progressions, and syncs everything across machines via Git.
The Problem
Claude Code sessions are ephemeral. Every new conversation starts from zero. Current memory mechanisms (MEMORY.md, CLAUDE.md) capture facts but lose the reasoning trail — why decisions were made, what dead ends were explored, how your understanding of a problem evolved through corrections and evidence. Patterns that emerge across sessions are never captured as reusable knowledge. And when you revisit a topic weeks later, you start from scratch instead of picking up where you left off.
What This Does
Six interconnected systems:
- Episodic Memory — Archive every session, generate structured summaries, enable full-text search across your entire conversation history
- Document Indexing — Index files in your knowledge repo (Markdown, code, PDFs, images) for full-text search alongside session history
- Skill Synthesis — Opus analyzes sessions and generates project-specific skills (structured prompts/playbooks) that capture recurring patterns, with auto-synthesis every N sessions
- Progressions — Track how your understanding of a topic evolves across a session or across sessions. A sequence of numbered documents that captures the full reasoning arc: initial assessment, deepenings, corrections, pivots, and synthesis. Corrections never decay — future sessions inherit what was wrong and why.
- User Patterns — Learn cross-project behavioral patterns from how you work (verification habits, investigation methodology, common corrections) and inject them as behavioral instructions into every session. Patterns get stronger with evidence from multiple sessions and projects.
- Preferences & Checkpoints — Store explicit user directives (
/remember always use bun) and save important discoveries during long sessions (pi-checkpoint) before they're lost to context compaction. Both persist in the knowledge repo and are injected into future sessions.
- Knowledge Repo — Git-backed per-project knowledge store that syncs across machines, versions skill evolution, stores progressions, and serves as the durable source of truth
Why Open Source
The tool is the plumbing. Your data is the value.
Your session history, skills, progressions, and accumulated project intelligence — that's private. It lives in your own Git repo, under your control, on your machines. Nobody sees it but you.
The tool that captures and retrieves that intelligence is commodity infrastructure. There's no competitive advantage in keeping the plumbing closed. The advantage is in what flows through it — and that's yours.
Open source means:
- You can audit exactly what runs. This system reads your Claude Code session transcripts and can send progression content to the Anthropic API. You should be able to inspect every line of code that touches your data.
- Community contributions make everyone's tool better. Better search, better context injection, better skill synthesis. Your private data stays private. The tool improves for everyone.
- Your company gets the upside with zero risk. If your organization uses this, their intelligence stays in their private knowledge repo. The open source tool just makes it more useful. There's nothing to protect by keeping it closed, and everything to gain from community improvements.
The value each team brings is their own accumulated intelligence — the reasoning chains, the corrections, the patterns extracted from their specific work. That's private by design. The tool that makes it searchable and injectable is better when it's open.
Architecture
