Multi-Project Setup

A single Graph Memory server can manage multiple projects simultaneously. Each project gets its own set of graphs, its own MCP endpoint, and its own configuration.

Configuration

Define your projects in graph-memory.yaml:

server:
  host: "127.0.0.1"
  port: 3000

projects:
  frontend:
    projectDir: "/home/dev/apps/frontend"

  api:
    projectDir: "/home/dev/apps/api"

  docs:
    projectDir: "/home/dev/apps/docs"

Start the server with the config file:

graphmemory serve --config graph-memory.yaml

Independent MCP Endpoints

Each project gets its own MCP endpoint:

Project	MCP Endpoint
frontend	http://localhost:3000/mcp/frontend
api	http://localhost:3000/mcp/api
docs	http://localhost:3000/mcp/docs

Connect each AI workspace to its corresponding project endpoint. This keeps context isolated — searching in the api project only returns results from the api codebase.

Independent Graphs

Each project maintains its own graphs:

• DocGraph — markdown documentation
• CodeGraph — source code symbols
• FileIndexGraph — all files and directories
• KnowledgeGraph — notes and decisions
• TaskGraph — tasks and work items
• SkillGraph — reusable procedures

Graph data is stored in each project's .graph-memory/ directory (or a custom path set via graphMemory):

projects:
  api:
    projectDir: "/home/dev/apps/api"
    graphMemory: "/home/dev/data/api-graphs"   # custom storage path

Per-Project Configuration

Each project can override server-level settings:

server:
  model:
    name: "Xenova/bge-m3"
    dtype: "q8"

projects:
  frontend:
    projectDir: "/home/dev/apps/frontend"
    chunkDepth: 3
    maxFileSize: 2097152    # 2 MB limit for this project
    graphs:
      code:
        include: "**/*.{ts,tsx,vue}"    # also index Vue files

  api:
    projectDir: "/home/dev/apps/api"
    author:
      name: "API Bot"
      email: "[email protected]"
    graphs:
      docs:
        include: "**/*.{md,mdx}"        # include MDX files

Project Selector in Web UI

The Web UI at http://localhost:3000 includes a project selector dropdown. Switch between projects to browse their graphs, search content, and manage tasks independently.

Memory-efficient indexing

When managing multiple projects, Graph Memory indexes them sequentially in three phases per project: docs → files → code. Embedding models are loaded lazily — the ONNX pipeline only initializes when the first embedding is needed — and each phase completes before the next begins. This means only one model is resident in memory at a time, keeping peak memory low regardless of how many projects you configure.

For setups with three or more projects using the default model, this reduces peak memory by up to ~3 GB compared to loading all models simultaneously.

Indexing Individual Projects

You can index a single project without starting the server:

graphmemory index --config graph-memory.yaml --project api

Or re-index all projects:

graphmemory index --config graph-memory.yaml --reindex

Per-Project Access Control

Restrict access to specific projects per user:

users:
  alice:
    name: "Alice"
    email: "[email protected]"
    apiKey: "mgm-key-abc123"
    passwordHash: "$scrypt$..."

projects:
  frontend:
    projectDir: "/home/dev/apps/frontend"
    access:
      alice: rw     # Alice has full access

  api:
    projectDir: "/home/dev/apps/api"
    access:
      alice: r      # Alice can only read