MemPalace/mempalace
mempalace
The best-benchmarked open-source AI memory system. And it's free.
GitHub metrics
- Stars
- 60K
- +135 this week
- Forks
- 7.6K
- 30-day growth
- —
- Open issues
- 365
| Day | GitHub stars |
|---|---|
| 2026-10-04 | 59403 |
| 2026-10-10 | 59488 |
| 2026-10-13 | 59538 |
What it does
MemPalace is a Python memory layer for AI assistants: it saves your past conversations and project files word for word on your own machine, then lets you or your agent search them by meaning.
The project deliberately avoids summarizing what it stores. Records are filed into a hierarchy the authors call a palace: each person or project gets a wing, subjects inside it become rooms, and the original text sits in drawers. A search can then stay inside one wing instead of scanning everything. ChromaDB is the default vector store, and the repository ships opt-in adapters for SQLite, a native Rust engine, Milvus, Qdrant and pgvector.
Day to day, you point the mempalace CLI at a code folder or at Claude Code session logs, and it indexes them. An MCP server exposes the palace to agents with 45 tools, and hooks for Claude Code, Codex CLI and Cursor save sessions on a schedule and before context compression. The README reports 96.6% recall at five on the LongMemEval benchmark using plain semantic search with no language model involved, and says the scripts to reproduce that figure are in the repository.
Why it matters
Coding agents forget everything between sessions, and many memory tools fix that by having a model condense chats into extracted notes. MemPalace keeps the raw text instead and searches it locally, so nothing is rewritten and, per the README, nothing goes off your machine unless you turn that on. Its storage interface is swappable, which lets a team move from embedded ChromaDB to a Postgres or Qdrant server without changing how agents call it.
Main features
- Stores conversations and files verbatim, then retrieves them with semantic search
- Wing, room and drawer structure lets searches stay scoped
- ChromaDB by default; Milvus, Qdrant, pgvector and SQLite adapters are opt-in
- MCP server with 45 tools, including knowledge-graph and agent diary operations
- Auto-save hooks for Claude Code, Codex CLI and Cursor
- Temporal knowledge graph stored in local SQLite
- Multi-arch Docker image for amd64 and arm64
- Optional embeddings from any OpenAI-compatible endpoint, such as Ollama or LM Studio
Who it's for
- Developers using Claude Code, Codex CLI or Cursor who want past sessions searchable
- Teams that want a shared memory server for several agents
- People who prefer local storage over a hosted memory service
- Builders who want to plug their own vector database into an agent memory layer
Strengths
- Keeps original text instead of model-written summaries
- Runs locally by default, with no API key needed for the core path
- Storage backend can be swapped without touching the rest of the system
- The README links benchmark scripts and committed per-question result files
Limitations
- Native Android/Termux installs are unsupported; the README points to a Debian PRoot container instead
- The GPU container image is not published, must be built locally and only builds on x86_64
- The first command that needs embeddings downloads the model, so it is slow and needs network
- Benchmark figures are the project's own; the README declines side-by-side comparisons with other memory tools
Getting started
The README recommends uv so the CLI lands in its own isolated environment; plain pip is meant for an activated virtualenv. A container image covers machines without a Python toolchain.
uv tool install mempalace
pip install mempalace
docker pull ghcr.io/mempalace/mempalace:latest
Requirements: Python 3.9+, a vector store (ChromaDB is bundled by default) and roughly 300 MB of disk for the default embedding model; no API key for the core local workflow.
Always follow the official instructions in the README on GitHub. This overview is our summary, not a copy.
Sources
- MemPalace/mempalaceRepositoryPrimarygithub.com
- MemPalaceOfficial sitePrimarymempalaceofficial.com
- mempalaceofficial.com/guide/getting-started.htmlWeb
Alternatives to mempalace
Graphify turns a software project and related material into a queryable knowledge graph. It parses code locally with tree-sitter, connects structures such as calls, imports and inheritance, and writes the result to graph.json. It can also produce an interactive graph.html viewer and a GRAPHREPORT.md report. Queries can search the graph, explain a concept or trace a path between concepts. Code analysis does not require an LLM or API key. Graphify distinguishes relationships extracted directly from source from relationships resolved by Graphify. It does not use embeddings or a vector store for its graph. Documents, PDFs and images can join the same graph, but their semantic processing uses an AI assistant or configured model backend rather than the code-only parser. The project also supplies a skill that can be registered with supported coding assistants including Claude Code, Cursor, Codex and Gemini CLI.
- Stars
- 126K
- Language
- Python
- License
- Apache-2.0
- Momentum
- +2.0% 7dEstablished
PixelRAG is a retrieval system that turns web pages, PDFs and images into screenshot tiles and searches those images instead of extracted text. It is the official code for the arXiv paper "PIXELRAG: Web Screenshots Beat Text for Retrieval-Augmented Generation", written by researchers at Berkeley SkyLab, BAIR and Berkeley NLP. The pipeline runs in stages. pixelshot renders a document into tiles with a headless Chromium browser driven over CDP. The pixelrag command then chunks and embeds the tiles with Qwen3-VL-Embedding and LoRA adapters fine-tuned on screenshot data, builds a FAISS index, and serves search through a FastAPI app on CPU or GPU. Each stage is its own pip extra, so a project installs only the parts it uses. The team also runs a hosted index of 8.28M Wikipedia pages behind a live API, which takes text or image queries; the README says it needs no API key. The renderer doubles as a Claude Code plugin called pixelbrowse, which has Claude screenshot a page and read the image, and as an npm package for opencode. Training code sits in a separate train/ project with its own pinned PyTorch and transformers versions.
- Stars
- 10K
- Language
- Python
- License
- Apache-2.0
- Momentum
- Established
turbovec is an in-process vector index for Python and Rust that stores embeddings at 2 or 4 bits per dimension and searches the compressed vectors directly. It is written in Rust with Python bindings and uses TurboQuant, a quantization method from Google Research. The README's headline figures: ten million 768-dimensional embeddings that need 31 GB as float32 fit in 4 GB, and search runs 4.4× faster than FAISS FastScan at 4-bit, at matching recall. Its benchmark table puts the 4-bit speedup at 3.8–5.1× across eight test cells on GCP ARM and x86 machines. These are the project's own benchmarks, and the scripts and raw JSON are published alongside them. No training step is needed, so a vector is searchable as soon as it is added, and removal takes microseconds. sync() writes only what changed since the last call. IdMapIndex keys vectors by your own uint64 ids and accepts an allowlist filter at query time. Drop-in stores exist for LangChain, LlamaIndex, Haystack and Agno. It is a library rather than a server, and it scans every vector instead of using a graph index.
- Stars
- 17K
- Language
- Rust
- License
- MIT
- Momentum
- Established
DigUp is a free macOS menu bar app that searches the files on your Mac by what is in them, using Google DeepMind's EmbeddingGemma 2 model running locally. You pick the folders it may read. DigUp then indexes pictures, screenshots, PDFs, documents, audio recordings and videos, and you find things by typing a plain description. A hit in a video or recording opens at the matching moment, and a PDF hit opens at the page. Exact strings such as invoice numbers and error codes are matched too, including text that Apple's Vision framework reads from screenshots. Code search is a separate index you switch on yourself: queries start with code: and results open at the line in editors such as Zed, VS Code, Cursor and Xcode. The README says nothing you index leaves the Mac. The app goes online to download the model once (865 MB) and for a daily update check that can be turned off, and it has no account and no telemetry. Internally it runs EmbeddingGemma 2 on llama.cpp with Metal and keeps vectors plus a keyword index in one SQLite file. It needs an Apple Silicon Mac on macOS 14 or later. The latest release is DigUp 0.6.0.
- Stars
- 1.5K
- Language
- Swift
- License
- MIT
- Momentum
- Emerging
Related AI tools
Earlyn
Business & productivity
Mac app that keeps an encrypted, searchable memory of your screen and meetings, with local MCP access for AI coding agents.
FreemiumFree tierAPIVerified