aakarim/OpenLore

What it solves

OpenLore addresses the problem of fragmented and stale documentation when multiple AI agents, teams, or repositories rely on the same knowledge. Instead of duplicating Markdown files across different projects—which leads to inconsistent versions of the truth—OpenLore provides a single, governed source of truth that agents can access and update in real-time.

How it works

It functions as an agent-native knowledge base that serves ordinary Markdown files as a virtual filesystem. It does not require a vector database or LLM ingestion pipeline. Instead, it provides several interfaces for agents to interact with the data:

• SSH/SFTP: Agents use familiar shell commands like ls, cat, grep, and find via a secure, in-memory Go interpreter (not a real OS shell).
• MCP (Model Context Protocol): Provides a standardized way for agents to retrieve knowledge.
• Web View: A human-friendly interface for browsing documentation.

Access is managed via identity-scoped views and Role-Based Access Control (RBAC), ensuring agents and users only see the docsets they are authorized to access. Writes are handled as atomic swaps and can be validated through plugins before being committed.

Who it’s for

• AI Agent Developers: Those needing a shared, live memory or artifact repository for teams of agents.
• Engineering Teams: Teams wanting to provide internal API docs, runbooks, and architecture notes to coding agents via a greppable interface.
• Enterprise Users: Organizations requiring governed knowledge contribution with permissions, review processes, and identity-scoped access.

Highlights

• Agent-native retrieval: Uses standard shell tools and composition patterns rather than bespoke APIs.
• Multi-transport support: Serves the same filesystem over SSH, MCP, and HTTP.
• Safe execution: Uses a pure-Go interpreter to prevent shell escapes and arbitrary process execution.
• Portable bundles: Ability to embed documentation directly into a self-contained binary or package as a desktop MCP extension.
• Governed writing: Supports read-only modes, scoped publishing, and atomic, conflict-aware writes.