How ginko works.
Git-native context management that captures what matters and surfaces it exactly when you need it.
The architecture
.ginko/
All context lives in your repo. Versioned with git. Works offline.
Chronological event stream. Every insight, decision, and achievement captured.
AI instructions, architecture decisions, and sprint context.
ginko start
Sub-second context loading. Event-based, cursor-tracked, zero re-parsing.
Claude Code, Cursor, or any AI that reads CLAUDE.md. Full context, zero re-explaining.
Core concepts
Every session captures events: insights, fixes, decisions, achievements. Stored as append-only JSONL. Fast to write, fast to read.
ginko log "Fixed auth flow" → captured instantly, surfaced on next session.
Ginko remembers where you left off. On start, it loads only new events since your last session. No re-processing, no stale context.
When you end a session, ginko creates a handoff summary. When you start fresh, your AI knows exactly what was happening.
ginko handoff "Completed auth, starting payments" → next session resumes smoothly.
Patterns that work get recorded. Gotchas that bite get flagged. Your AI learns from your codebase, not just documentation.
.unref() on timers" → surfaced automatically when relevant.
The daily flow
START
Run ginko start. Context loads in under 2 seconds. Your AI knows where you left off.
WORK
Code as normal. Log insights with ginko log. Context captures automatically.
COMMIT
Commit your code. Context is versioned alongside. Session logs included if you want.
HANDOFF
Optional: ginko handoff creates a summary. Tomorrow's session starts instantly.
What's in .ginko/
The ginko collaboration graph
Local context is powerful. Cloud-synced context is transformative. Pro
Neo4j-powered knowledge graph with vector embeddings
Events sync to the cloud as you work. Team members see updates instantly. No manual exports, no stale context.
Ask questions in natural language. "How do we handle authentication?" returns relevant ADRs, patterns, and recent decisions.
AI navigates relationships: Pattern → related ADR → implementing sprint → relevant gotchas. Context arrives pre-connected.
See what teammates discovered. Patterns from Sarah's session surface in yours when relevant. Knowledge compounds across the team.
How it works
Events logged locally in JSONL format. Works offline, zero latency.
Events batch-sync to cloud graph every 5 minutes or on threshold.
Each event embedded for semantic similarity. Relationships auto-detected.
Queries traverse graph + vectors. Most relevant context surfaces first.
Integrations
Reads CLAUDE.md automatically. Full context on every session. First-class support.
Works via .cursor rules. Same context, different editor. Seamless switching.
Native integration. Context versions with code. Branching, merging, diffing all work.
Ready to try it?
Install in 30 seconds. No account required for local use.