AI system

Private knowledge & memory service

A self-hosted retrieval and memory layer that keeps project knowledge searchable across tools, with scoped access, provenance tracking and no data leaving the network.

  • Retrieval
  • Local models
  • MCP
Category
Local AI & retrieval system
Scope
Indexing, scoped memory, provenance & MCP server
Core stack
Retrieval, Local models, MCP

The problem

Project notes, decisions and technical context end up split across editors, chats and docs. Hosted AI memory tools usually mean uploading private material to someone else's cloud, and most of them cannot tell you where a recalled fact came from or whether it is still current.

We built a memory service that runs on our own hardware, keeps projects separated and tracks where every fact comes from.

What we built

A self-hosted search and memory service that indexes documents and short facts locally, and exposes them to editors and agents over MCP.

  • Project scopes Records are grouped by project and scope, so a tool only sees the material it is allowed to read.
  • Source tracking and history Facts keep a link to their source document. When something changes, the old entry is marked superseded instead of quietly overwritten.
  • Local search Indexing and embeddings run on local hardware. Nothing leaves the network.
  • MCP interface Editors, coding agents and scripts all talk to the same endpoint, so knowledge stays shared across tools.

Why we built it this way

If an assistant cites a fact, you should be able to check where it came from and fix it when it goes stale. Running it as a normal database with explicit scopes and history is more work up front, but you can actually trust what comes back.

Need private AI or retrieval in your stack?

Tell us what the system needs to know and where the data has to live. We will tell you honestly whether a local setup makes sense and what it would take.

hello@bigfriendlycompany.eu