
mempalace
Local-first verbatim memory store for AI assistants, with semantic search over your chats
The repository was last updated 2 days ago and is published under MIT.
- Pricing
- Open source
- Platforms
- CLI, macOS, Linux
- GitHub stars
- 60.6K
- Last commit
- 2 days ago
- Language
- Python
- License
- MIT
Similar MCP servers
About mempalace
mempalace is an open-source MCP server for developer tools. It is available on CLI, macOS, Linux, Windows, self-hosted and API.
The README says MemPalace keeps conversation history and project notes as verbatim text and retrieves them with semantic search, without summarizing or paraphrasing what was stored. Content is organized into wings for people and projects, rooms for topics, and drawers that hold the original text, so searches can be scoped instead of run across a flat collection.
It runs on the machine itself: ChromaDB is the default vector store, a SQLite database backs a temporal knowledge graph, and embedding runs locally with a choice of two models. The MCP server exposes its palace operations to clients such as Claude Code, and auto-save hooks for Claude Code, Codex CLI and Cursor file new material during a session.
It suits developers and power users of AI coding assistants who want past sessions and project files recallable across runs. The README also lists alternative backends (Qdrant, pgvector, Milvus, and others) and a Docker image for running the MCP server without a local Python setup.
What it does
- Verbatim drawer storage
- Stores original chunks of text unchanged and indexes them so they can be located later.
- Scoped semantic search
- Organizes content into wings, rooms and drawers so queries can be limited to a person, project or topic.
- Local embeddings
- Generates embeddings on the machine with a choice of MiniLM or EmbeddingGemma, with an optional OpenAI-compatible endpoint.
- Temporal knowledge graph
- Records entity relationships with validity windows, backed by local SQLite.
- Claude Code and Cursor hooks
- Auto-save hooks file conversation content before context compression and restore recall at session start.
- Pluggable storage backends
- Ships ChromaDB by default and supports other backends such as Qdrant and pgvector through configuration.
Good for
- Recalling decisions and context from earlier Claude Code sessions in a new session
- Searching past project notes and files by meaning rather than exact keywords
- Keeping a private, offline-capable memory store for AI assistants
- Tracking which facts about a project or person were true at a given time
Pricing
Open source
Free and open source under the MIT license; self-hosted.
Prices change. Check mempalaceofficial.com before you buy.
mempalace FAQ
Does MemPalace send my data to the cloud?
The README says nothing leaves your machine unless you opt in, and the palace lives in a single local directory. Using an OpenAI-compatible embedding endpoint is optional and stays on your network when the endpoint is local or on your LAN.
Which license does it use?
MemPalace is released under the MIT license.
Does it need an API key?
The README states that no API key is required for the core benchmark path. Optional LLM reranking uses a model of your choice.
How do I install it?
The README recommends installing the mempalace package with uv tool install, pipx, or pip inside a virtual environment, then running mempalace init.
Is mempalace free?
Yes. mempalace is open source, so you can use and self-host it for free.
What are the best alternatives to mempalace?
The closest alternatives to mempalace are Memro, Total Agent Memory, ChatCrystal, Hypermnesia MCP and Memoria — Deterministic AI Memory & Code Intelligence MCP.
Where this comes from
- Description: written from the project's README and the project's website.