# Connapse

An open-source, self-hosted knowledge backend that gives AI agents persistent, searchable memory over uploaded documents or existing storage, through an MCP server, REST API, or CLI, built on .NET.

Connapse is a self-hosted knowledge backend addressing the problem that an AI agent otherwise starts every new conversation with no memory of past research or access to previously ingested documents. A user uploads documents or points it at existing Amazon S3 buckets or local filesystems, and agents then query and build their own research corpus through 11 MCP tools, a REST API, or a CLI, backed by hybrid vector-and-keyword search.

It runs as container-isolated services, deployable with Docker Compose in about 60 seconds, using PostgreSQL with the pgvector extension for storage and MinIO for object storage, and supports local embeddings through Ollama as well as configurable embedding providers, chunking parameters, and search settings that can be changed at runtime without restarting. A separate command-line client and a Claude integration through the built-in MCP server are also provided.

It is built on .NET 10.

Website: https://github.com/Destrayon/Connapse