Our gradio app for MCP-powered memory.

A full RAG pipeline using Vectara, with an optional persistent memory system using a knowledge graph, built with Convex backend. This system allows for storing information about entities, their relations, and observations across multiple interactions.

git clone https://github.com/TanGentleman/YouTwo
cd YouTwo
python -m venv .venv
source .venv/bin/activate
pip install -r requirements-extra.txt
python scripts/run_gradio_app.py

cd YouTwo
modal deploy deployment/modal_app.py

The knowledge graph memory server implements:

1. Entity Management: Create, update, and delete entities with associated metadata
2. Relationship Tracking: Track how entities relate to each other
3. Journal System: Record timestamped entries that can be analyzed for entity extraction
4. Semantic Search: Find semantically similar content using vector embeddings
5. HTTP API: External integration with the knowledge graph

• Node.js (v14 or later)
• npm

1. Clone the repository:

git clone https://github.com/TanGentleman/YouTwo
cd YouTwo/backend

2. Install dependencies:

npm install

3. Start the development server (You will be prompted to login to Convex):

npx convex dev

Expose all our backend functions easily using MCP or even http requests. See: https://docs.convex.dev/client/react/deployment-urls https://docs.convex.dev/ai/convex-mcp-server

The system provides both a JavaScript API and HTTP endpoints for managing the knowledge graph.

await createEntities({
  entities: [
    {
      name: "John_Smith",
      entityType: "person",
      observations: ["Speaks fluent Spanish", "Graduated in 2019"]
    }
  ]
});

await createRelations({
  relations: [
    {
      from: "John_Smith",
      to: "Anthropic",
      relationType: "works_at"
    }
  ]
});

const graph = await readGraph();

const results = await searchNodes({ query: "Spanish speaker" });

The system also exposes HTTP endpoints for external integration:

• POST /entities - Create entities
• POST /relations - Create relations
• POST /observations - Add observations to entities
• GET /graph - Read the entire knowledge graph
• GET /search?q=query - Search nodes by query
• GET /nodes?names=name1,name2 - Get specific nodes by name
• DELETE /entities - Delete entities
• DELETE /observations - Delete observations
• DELETE /relations - Delete relations

The system is built using Convex for the backend database and API. The codebase is structured as follows:

• backend/convex/ - Convex backend code
  • schema.ts - Database schema definition
  • entities.ts - Entity management functions
  • relations.ts - Relation management functions
  • knowledge.ts - Knowledge graph query functions
  • journals.ts - Journal entry management
  • embeddings.ts - Vector embedding functions
  • distiller.ts - Knowledge extraction from journals
  • operations.ts - Operation logging utilities
  • http.ts - HTTP API endpoints

MIT