OpenContracts: Version Control for Knowledge and AI Agents

What changed

OpenContracts, developed by Open-Source-Legal, is presented as a platform for creating a programmable citation graph from document repositories. This system aims to serve as a "ground truth layer" for collaborative work between humans and AI agents. The core functionality involves ingesting documents, automatically detecting and resolving citations to build a navigable graph. This graph can then be accessed through multiple interfaces: a GraphQL and REST API for applications, a Model Context Protocol (MCP) server for AI agents, and a React UI for human interaction. The platform supports self-hosting and is released under an MIT license, targeting teams working with large document sets.

Key features include:

• Automated Citation Graph Generation: Documents are processed to identify and link statutory and other citations, creating a structured representation of knowledge. Unresolved citations are tracked as a backlog.
• Multi-Surface Access: The same underlying graph data is exposed via a GraphQL/REST API, an MCP endpoint for AI agents (compatible with tools like Claude and Cursor), and a user interface.
• AI Agent Integration: Developers can spin up document- or corpus-scoped AI agents in Python, which can stream responses grounded in the citation graph. The MCP server allows external agents to search the corpus, traverse citation edges, and propose annotations.
• Structured Extraction: The platform supports defining "fieldsets" – natural language queries for data extraction – which can be run across entire corpuses, with results available in a spreadsheet-like grid and subject to human approval.
• Pluggable Pipeline: Components for parsing, embedding, and thumbnailing are swappable, allowing customization for different document formats and downstream processing.
• Version Control for Knowledge: Corpuses are treated as version-controlled collections, akin to Git repositories, enabling branching, sharing, and restoration of previous states. Human annotations are treated as the ground truth for building and refining the citation graph.

Why it matters for builders

OpenContracts addresses a fundamental challenge in AI: enabling agents to reason effectively over complex, interconnected information. Instead of relying solely on large context windows, it provides a structured knowledge substrate that AI agents can actively navigate. This allows for more reliable grounding of AI-generated answers and facilitates deeper analysis of document relationships, which is critical for domains like legal, research, and engineering.

Practical impact

Builders can leverage OpenContracts to create AI-powered tools that go beyond simple document retrieval. By integrating with the GraphQL/REST API or the MCP server, developers can build custom applications or agents that understand and utilize the citation graph. This could involve developing advanced research tools, automated compliance checkers, or sophisticated knowledge management systems. The ability to define custom extractors and pluggable components offers flexibility for tailoring the system to specific data formats and analytical needs. The version control aspect for knowledge also allows for iterative development and auditing of AI-driven insights.

Caveats and source limits

The provided source material describes the functionality and architecture of OpenContracts but does not include specific benchmark results, pricing details for enterprise use, or a definitive release date beyond a "latest release: v3.0.0.b4" mention. The project is MIT-licensed and self-hosted, implying potential infrastructure costs and management overhead for users. The effectiveness of the AI agents and the accuracy of the citation graph generation will depend on the quality and nature of the ingested documents and the human annotations provided.