Open-Source-Legal's 'cite' Project: Version Control for Knowledge and AI Agents

The 'cite' project by Open-Source-Legal, previously known as OpenContracts, is a Python-based initiative focused on establishing a "ground truth layer" for knowledge. It functions as a version control system for documents and their interconnections, designed for both human and AI agent collaboration. The core idea is to transform document repositories into an open citation graph where documents are nodes and citations are edges. This graph allows AI agents to navigate relationships between documents, cite specific spans of text, and propose new annotations, which humans can then review and accept.

The project emphasizes the importance of a structured substrate for AI agents, arguing that current methods often lead to agents hallucinating relationships or failing to resolve references. 'cite' aims to provide this substrate through a GraphQL and REST API for human and application interaction, and a Model Context Protocol (MCP) endpoint for agents. The underlying engine includes features for annotation, corpus management, AI agents, MCP server, and vector search.

Key features highlighted in the latest beta release (v3.0.0.b4) include Auth0 authentication for Django Admin, runtime-configurable pipeline settings with encrypted secrets storage, personal corpus creation for users, and enhanced document processing pipeline hardening with retry mechanisms. It also introduces bifurcated conversation permissions, corpus forking improvements, corpus-scoped MCP endpoints, and multimodal embedding support. Frontend modernization efforts include a unified upload modal, redesigned corpus list view, and real-time updates in the Extract View. Security enhancements address WebSocket agent permissions, JWT error message hardening, sensitive data redaction in logs, and IDOR prevention. Other notable changes include LlamaParse integration, real-time notification system, and bulk document management. The project is licensed under MIT.