project-mind 0.2.0

Persistent AI context management for multi-tool development — syncs Claude, Gemini, Cursor, Codex and Copilot transcripts into a single LLM-tuning file.

What is mind?

mind is a single file memory layer for AI-assisted development. It reads your conversations from Claude Code, Gemini, Cursor, Codex, and Copilot after every commit and synthesizes them into one token efficient mind.md that every tool can load as context.

I built this because I kept hitting the same wall across projects — every AI tool has its own rules, its own memory, its own prompt format, and none of them talk to each other. You can configure behaviors per tool (rules and skills etc), but there's no shared, consistent context that travels with the project. So I'd find myself re-explaining the same architecture decisions, the same gotchas, the same "we tried that and it didn't work" notes to a different model every other day.

mind is the simplest fix I could think of: one file, written and updated by LLM after each commit, aggregating context from every tool you use into one place.

Install

pip install project-mind

Quick start

cd your-project
mind init      # creates _mind/, installs git hook, writes mind.toml
mind sync      # extract transcripts and rebuild mind.md
mind status    # show tracked tools and last sync

The git hook runs mind sync automatically after every commit.

Supported tools

Tool	Source
Claude Code	~/.claude/projects/{slug}/*.jsonl
Gemini	~/.gemini/tmp/{project}/chats/*.json
Cursor	~/.cursor/projects/{slug}/agent-transcripts/
Codex	~/.codex/sessions/**/rollout-*.jsonl
Copilot	VS Code workspaceStorage/{hash}/state.vscdb

How it works

1. After every git commit, the hook runs mind sync in the background.
2. mind sync finds transcripts newer than the last sync and extracts the user + assistant text.
3. It calls your configured LLM with the extracted content and the current mind.md.
4. The LLM rewrites mind.md — compressing old entries, adding new ones, preserving the sections you care about.

mind.md structure

## behavior     user corrections — never compressed
## context      project state — rewritten each sync
## active       in-flight tasks
## decisions    architectural choices
## lessons      what worked, what didn't
## history      compressed timeline

Configuration

# mind.toml
[project]
name = "my-project"

[llm]
provider = "claude"   # claude | gemini | codex

[tools]
enabled = ["claude", "gemini", "cursor", "codex", "copilot"]

[limits]
max_messages_per_sync = 150
mind_max_lines = 150

License

MIT