Skip to content
Get packs

MemPalace

Local-first AI memory that stores your conversations verbatim and searches them semantically.

GitHub Download v3.11.0
Stars
60.2k
Forks
7.6k
Licence
MIT
Language
Python
Last update
6 days ago
Type
App

What it is

MemPalace keeps a record of your conversation history as exact, unedited text and lets you search it by meaning. It organizes that material into wings (people and projects), rooms (topics), and drawers (original content) so searches can be narrowed. According to the README, nothing leaves your machine unless you opt in, and it is designed to act as memory for AI coding assistants through MCP.

  • Mine a project directory into a palace so you can search it later with mempalace search
  • Connect the MCP server to Claude Code so the agent can search past conversations before answering
  • Run the MCP server or CLI in Docker with a persistent volume for storage
  • Scope searches to a particular person, project, or topic using the wing and room structure

Great for

  • Developers who use Claude Code or another MCP-capable coding agent and want it to recall past sessions
  • People who want their conversation history kept locally rather than summarized by a third party
  • Users comfortable working in a terminal and with Python tools like uv or pipx
  • Teams interested in the shared-brain or team server option described in the README

Not ideal for

  • Anyone wanting a polished point-and-click app, since the README centers on CLI, MCP, and Docker setup
  • Android users on Termux, since the README says native Termux installation is not currently supported
  • People on ARM Linux or Apple Silicon who need the GPU Docker image, since the README says it is x86_64-only
  • Users who need guaranteed, fully explained benchmark claims, since the README gives a headline score without detailed methodology in the excerpt
Step-by-step guide: how to use MemPalaceInstall, first run, examples, tips and FAQs
MemPalace alternativesCowAgent vs MemPalaceMemPalace vs MCP for Unity

How to get it

  1. 1

    Install with a package manager

    You'll need: Python (version not specified in the README excerpt) · uv, pipx, or pip · Docker (optional, for the container route)

    Install the CLI with uv (recommended in README)

    uv tool install mempalace

    Initialize a palace for a project folder

    mempalace init ~/projects/myapp
  2. 2

    Run it

    • Install the CLI with uv tool install mempalace or pipx install mempalace
    • Run mempalace init on a project folder to create a palace
    • Mine content into the palace using the mine command shown in the README
    • Search with mempalace search, or wire the MCP server into your coding agent
  3. 3

    Get the most out of it

    • Wire up auto-save hooks for Claude Code to avoid the 30-day session expiry the README warns about
    • Use the wing and room structure to scope searches rather than querying the whole corpus
    • Mount your transcript folder read-only into Docker so the MCP server can mine it without write access
    • Use the agent-guided setup via npx skills add MemPalace/mempalace if you prefer the agent to handle configuration
    • Pick a non-default backend such as qdrant or pgvector via MEMPALACE_BACKEND if you need a server-based store

Downloads

Source code (.zip)develop branch · straight from GitHubRelease v3.11.08 files incl. mempalace_native_core-3.11.0-cp39-abi3-macosx_10_12_x86_64.whlOfficial website & docsmempalaceofficial.com/

Strengths

  • Stores text verbatim rather than summarizing or paraphrasing, per the README
  • Pluggable storage backends, with ChromaDB as the default and no configuration needed
  • Released under the MIT license and the repository reports a large star count

Trade-offs

  • Setup involves several moving parts (CLI, MCP configuration, Docker volumes, embedding model downloads), so it is not a one-click install
  • The README warns that Claude Code sessions expire in 30 days without auto-save hooks wired up, which means extra setup is needed to avoid data loss
  • The README's headline benchmark figure is a claim the excerpt does not fully explain, so readers should check the linked documentation

How it compares

ProjectStarsTypeLevelUpdatedBest for
mempalace (this)60.2kAppintermediate6 days agoDevelopers who use Claude Code or another MCP-capable coding agent and want it to recall past sessions
paperclip110kAppadvanced4 days agoDevelopers already running several AI coding agents who want one place to track them
nanobot49.8kAppintermediate4 days agoPeople who want a self-hosted personal assistant they can inspect and customize
CowAgent47.8kAppintermediate4 days agoPeople who want a self-hosted personal assistant reachable from their everyday chat apps
Horizon11.6kAppintermediate4 days agoPeople who want a personalized daily news digest they run themselves
cc-switch151kAppbeginner4 days agoDevelopers who use several AI coding agents and switch between API providers often

Stats synced from GitHub. The assessment above was drafted with AI from the project's README on 2026-10-29 and may miss recent changes — check the README before relying on a command. MIT licence: free to use, modify and distribute, including commercially, as long as you keep the copyright notice.