Install with Docker Compose

Install mnemos on your own server with Docker Compose. This example uses PostgreSQL, Ollama, an MCP server and an HTTP API. If you only connect an AI assistant over MCP, you can omit the api service.

Requirements

• Docker with the Compose plugin;
• 4 GB of RAM: the embedding model takes about 2 GB;
• a TLS reverse proxy if the server is reachable from outside. The containers listen on 127.0.0.1 only.

1. Create docker-compose.yml

name: mnemos

services:
  postgres:
    image: pgvector/pgvector:pg17
    environment:
      POSTGRES_USER: mnemos
      POSTGRES_PASSWORD: ${MNEMOS_DATABASE_PASSWORD}
      POSTGRES_DB: mnemos
    volumes:
      - pgdata:/var/lib/postgresql/data
    healthcheck:
      test: ["CMD-SHELL", "pg_isready -U mnemos"]
      interval: 5s
      retries: 20

  # Local model for search by meaning.
  ollama:
    image: ollama/ollama:latest
    volumes:
      - ollama:/root/.ollama
    entrypoint: ["/bin/sh", "-c"]
    command: >
      "ollama serve &
       until ollama list >/dev/null 2>&1; do sleep 1; done;
       ollama pull bge-m3;
       wait"
    healthcheck:
      test: ["CMD-SHELL", "ollama list | grep -q bge-m3"]
      interval: 10s
      retries: 60
      start_period: 30s
    restart: unless-stopped

  mnemos:
    image: cr.yandex/crpo20ag926fr5ijfri7/mnemos-mcp:latest
    env_file: .env
    depends_on:
      postgres:
        condition: service_healthy
      ollama:
        condition: service_healthy
    ports:
      - "127.0.0.1:8090:8090"
    restart: unless-stopped

  # API for apps and scripts.
  api:
    image: cr.yandex/crpo20ag926fr5ijfri7/mnemos-api:latest
    env_file: .env
    depends_on:
      mnemos:
        condition: service_started
    ports:
      - "127.0.0.1:8091:8090"
    restart: unless-stopped

volumes:
  pgdata:
  ollama:

2. Create .env

Save it next to docker-compose.yml. Enter your database password and authentication addresses.

# Database
MNEMOS_DATABASE_HOST=postgres
MNEMOS_DATABASE_USER=mnemos
MNEMOS_DATABASE_PASSWORD=change-me
MNEMOS_DATABASE_NAME=mnemos
MNEMOS_DATABASE_SSLMODE=disable

# Embeddings
MNEMOS_EMBEDDINGS_BASE_URL=http://ollama:11434/v1
MNEMOS_EMBEDDINGS_MODEL=bge-m3
MNEMOS_EMBEDDINGS_DIMENSIONS=1024

# Authentication
MNEMOS_AUTH_ISSUER_URL=https://auth.example.com
MNEMOS_AUTH_PUBLIC_URL=https://mnemos.example.com

3. Start mnemos

docker compose up -d

On the first start, wait for the model to download — about 2 GB.

Check the server status:

curl -s http://127.0.0.1:8090/healthz
docker compose logs --tail 20 mnemos

Embeddings: locally or through a provider

Locally (the default)

The example uses Ollama with the bge-m3 model, which needs about 2 GB of memory. Text for search is processed on your server.

MNEMOS_EMBEDDINGS_BASE_URL=http://ollama:11434/v1
MNEMOS_EMBEDDINGS_MODEL=bge-m3
MNEMOS_EMBEDDINGS_DIMENSIONS=1024

Through a provider

For an external service with a /v1/embeddings API, set its address, key and model. Example for OpenAI:

MNEMOS_EMBEDDINGS_BASE_URL=https://api.openai.com/v1
MNEMOS_EMBEDDINGS_API_KEY=sk-...
MNEMOS_EMBEDDINGS_MODEL=text-embedding-3-large
MNEMOS_EMBEDDINGS_DIMENSIONS=1024
MNEMOS_EMBEDDINGS_REQUEST_DIMENSIONS=true

Use 1024 dimensions. Keep MNEMOS_EMBEDDINGS_REQUEST_DIMENSIONS=true for this example; Ollama does not need that parameter.

Authentication

Set two addresses:

• MNEMOS_AUTH_ISSUER_URL — your OAuth server's address.
• MNEMOS_AUTH_PUBLIC_URL — mnemos's public address, such as https://mnemos.example.com.

The server will not start without them: there is no unauthenticated mode.

Upgrading

docker compose pull
docker compose up -d

After startup, connect your assistant. See configuration for other settings.