Docker

The Community release is designed to be built with Docker Compose, which runs the Go server with Pebble storage and the local embedding runtime, and installs the embedding model on first start. Use Docker Compose for the recommended setup.

Build the image

From a Community checkout:

docker build -t remem:0.5.0 -f docker/Dockerfile .

Install the embedding model

The image does not include the embedding model. Install it into a host directory once, before the first start. This downloads about 2.4 GB from Hugging Face, with no account needed, and checks every file:

mkdir -p ./remem-data ./remem-model
sudo chown 10001:10001 ./remem-data ./remem-model

docker run --rm \
  -v "$PWD/remem-model:/var/lib/remem-model" \
  --entrypoint remem-install-model \
  remem:0.5.0

To install from Docker Hub instead of Hugging Face, run the model image:

docker run --rm -v "$PWD/remem-model:/var/lib/remem-model" \
  rememorg/remem-model:pplx-embed-context-v1-0.6b-b42df969

Running either again only verifies the files, so it is safe to repeat.

Run with persistence

docker run -d \
  --name remem \
  --restart unless-stopped \
  -p 4545:4545 \
  -v "$PWD/remem-data:/var/lib/remem" \
  -v "$PWD/remem-model:/var/lib/remem-model:ro" \
  -e REMEM_SERVER_API_KEY=your-secret-key \
  -e REMEM_SERVER_ENV=production \
  remem:0.5.0

Pebble keeps the corpus in /var/lib/remem. The bind mount makes it visible to ordinary host backup tools and keeps it across container replacement. The server reads the model from /var/lib/remem-model and refuses to start if it is missing.

Verify the container

curl http://localhost:4545/api/v1/ready
docker logs remem --follow

The HTTP API and Streamable HTTP MCP endpoint are available on port 4545. For local MCP clients, run the remem-mcp companion binary and point it at http://localhost:4545/mcp.