PostgMem

What is PostgMem?

PostgMem is a .NET-based MCP (Model Context Protocol) server that provides vector memory storage using PostgreSQL with the pgvector extension. It allows AI agents to store, retrieve, and search through memories using vector embeddings for efficient similarity search.

How to use PostgMem?

Ensure .NET 9.0 SDK, a PostgreSQL database with pgvector installed, and a text embedding API (default is Ollama) are available. Create the required memories table, configure the .env file with connection string and embedding settings, then run dotnet run in the PostgMem directory. The server starts at http://localhost:5000 by default.

Key features of PostgMem

• Store structured memories with vector embeddings
• Retrieve memories by ID
• Semantic search using vector similarity
• Filter search results with tags
• MCP integration for easy use with AI agents

Use cases of PostgMem

• Store and retrieve conversation memories with tags
• Store and retrieve document memories with tags
• Semantic search over memories with similarity threshold and tag filtering
• Delete specific memories by ID

FAQ from PostgMem

What are the prerequisites to run PostgMem?

You need .NET 9.0 SDK, a PostgreSQL database with the pgvector extension installed, and a text embedding API (the default configuration uses Ollama).

How do I configure the embedding API and database connection?

Set the ConnectionStrings__Storage, Embeddings__ApiUrl, and Embeddings__Model environment variables in a .env file. The embedding API URL defaults to http://localhost:11434/.

What MCP tools does PostgMem provide?

It provides four tools: Store (store a new memory), Search (search by similarity), Get (retrieve by ID), and Delete (delete by ID). Each accepts specific parameters documented in the README.

What database schema is required?

A memories table must be created with columns: id (UUID), type (TEXT), content (JSONB), source (TEXT), embedding (VECTOR(384)),