Practical overview
When it fits
Use Mempalace when an AI application needs persistent, verbatim conversation recall and a local-first storage approach with a ChromaDB default and pluggable retrieval layer.
Who it is for
- Developers building self-hosted AI assistants or agents.
- Teams prototyping local-first chat applications with persistent recall.
- Users who need a CLI, Docker, or MCP-based memory service.
- Developers experimenting with pluggable AI-memory backends.
Documented uses
- Store conversation history without summarization or paraphrasing.
- Retrieve relevant memories through semantic search.
- Add persistent memory to a self-hosted assistant.
- Expose memory capabilities through an MCP-based integration.
- Prototype different storage backends behind the memory layer.
Requirements & limits
The repository is public, uses the MIT license, and documents installation through uv, pipx, a Python virtual environment, or Docker. Native Termux installation is not currently supported; Android ARM64 use is documented through a Debian PRoot container.
- Whether verbatim storage is preferable to summarized or paraphrased memory.
- Whether local-first operation and zero API calls fit the application design.
- Whether ChromaDB or another pluggable backend suits the retrieval setup.
- Whether the deployment environment can support Python dependencies, Docker, or MCP.
- Whether persistent storage can be mounted, especially for Docker deployments.
Check current requirements in the official documentation. Source-based guidance does not mean OpenViST has run or independently tested this project.
Install Mempalace with uv and initialize a project
Install the Mempalace CLI in an isolated environment and initialize a project directory.
Before you start
- A Python-capable environment with uv available.
- A project directory path such as ~/projects/myapp.
Steps
Run the command below to place the CLI in an isolated environment on the PATH.
uv tool install mempalaceRun
mempalace init ~/projects/myappto initialize the project.
Expected outcome
The Mempalace CLI is installed in an isolated environment and an initialized project directory is created.
Guide limitations
- The documented steps cover installation and initialization, not a subsequent memory-ingestion or retrieval workflow.
Run the Mempalace MCP server in Docker
Start the Dockerized MCP server with persistent Mempalace data.
Before you start
- Docker.
- A Docker volume named mempalace-data or an equivalent persistent volume.
- The image ghcr.io/mempalace/mempalace.
Steps
Pull the image with the command below.
docker pull ghcr.io/mempalace/mempalace:latestRun the command below.
docker run -i --rm -v mempalace-data:/data ghcr.io/mempalace/mempalaceFor MCP clients that need source files, add a read-only bind mount and use the resulting container path when mining.
Expected outcome
The MCP server runs over standard input with palace data, configuration, and cached embeddings stored under `/data`.
Guide limitations
- The first embedding-dependent call downloads a model and requires network access.
- Mounted source directories must be readable by the container's uid 1000 on Linux.
- MCP clients should use absolute host paths;
~and$HOMEare not expanded by every client.
Sources & review
Source record details
- Source availability
- Active
- Sources checked
- 1 October 2026
- Original website title
- MemPalace/mempalace · Official homepage title · retained in its source language