Everything is readable Markdown, JSON and a log — the SQLite index is derived and can be rebuilt with one command, so nothing important is trapped in a database. The two scopes are the design choice that matters: a decision about this repository and a preference about how you like explanations are different things and get stored separately.
A local-first continuity tool for AI coding sessions. It captures what you were doing — task, git state, decisions — into files under `.mindswap/`, and serves that same context to AI clients over MCP so the next tool picks up cleanly instead of starting cold.
- A repository set up and existing AI context imported — `init`; the current task, git state and decisions captured — `save`
- Setup health and stale context checked — `doctor`; a clean briefing to start from — `resume`
- Project memory searched with citations — `ask`; blockers, assumptions, questions and resolutions managed — `memory`
- Two memory scopes: repo memory in `<repo>/.mindswap/` and personal memory in `~/.mindswap/`, with `--global` for preferences that should follow you across projects
- The local SQLite search index rebuilt from the file-based memory — `reindex`; continuity state shared across machines — `sync`
- The same context exposed to AI clients — `mcp` and `mcp-http`, with `mcp-install` to wire up a client
- Generated context for Claude Code, Cursor, Copilot, Codex, Windsurf, Cline, Roo, Aider, Amp, Gemini CLI and AGENTS.md-based workflows
No account and no key — memory lives in plain files on your machine. Install with `npm install mindswap --save-dev`, then `npx mindswap init`. The MCP entry runs the `mindswap` package with the positional argument `mcp` over stdio; version 3.3.0, MIT licensed. Repository state sits in `.mindswap/` as `HANDOFF.md`, `state.json`, `decisions.log`, `memory.json`, `config.json`, plus `branches/` and `history/`.
One command — npm install mindswap --save-dev