<2-3 sentences summarising what the project addresses and solves, the target audience, and the benefit.> rac-core (Lore) addresses the challenge of coding agents ignoring established team decisions by providing a deterministic system of record. Storing product knowledge—requirements, decisions, designs, and prompts—as versioned Markdown files in a repository, it serves this knowledge read-only to agents via the MCP protocol. This ensures agents cite and adhere to team decisions, preventing rework and grounding them in organizational context, benefiting teams heavily reliant on AI coding assistants.

How It Works

The system leverages typed Markdown artifacts within a Git repository, each with a frontmatter envelope. The rac-core engine deterministically classifies and validates these artifacts against schemas. Retrieval is designed to be deterministic and reproducible, focusing on returning exact, current decisions rather than fuzzy matches. Crucially, write-time enforcement via rac validate and rac gate in CI prevents malformed artifacts or references to superseded decisions from being merged. The core engine operates air-gapped, making no LLM or network calls during operation, with only an optional, consent-gated usage ping.

Quick Start & Requirements

• Installation: Install the core engine and CLI via pip: pip install rac-core. Additional ingest or explorer features can be installed with extras like [ingest-all] or [explorer]. uv tool install rac-core is also supported.
• Scaffolding: Initialize your project and first artifact with rac quickstart.
• Agent Connection: Connect agents like Claude Code using claude mcp add lore -- rac mcp. Configuration details for Claude Desktop and Cursor are provided.
• Prerequisites: Requires Python 3.11+.
• Documentation: Full documentation is available at https://itsthelore.github.io/rac-core/.

Highlighted Details

• Deterministic Knowledge Grounding: Unlike RAG or agent memory, Lore provides exact, current decision retrieval, acting as a read-only source of truth.
• CI Enforcement: rac validate and rac gate commands enforce artifact integrity and decision adherence in continuous integration pipelines before code merges.
• Multi-Format Ingestion: Supports importing knowledge from various sources including Confluence, Notion, DOCX, HTML, PDF, PPTX, and XLSX using rac-import or rac-ingest skills.
• Flexible Export Options: Knowledge can be exported as a single HTML file (Portal), conformant Open Knowledge Format (OKF) bundles, JSONL for memory/RAG backends, or graph data.
• MCP Protocol: Enables seamless integration with supported AI coding agents.

Maintenance & Community

The project welcomes contributions, ideas, and experiments. Specific details regarding notable contributors, sponsorships, or dedicated community channels (like Discord/Slack) are not explicitly detailed in the README.

Licensing & Compatibility

The project is licensed under the Apache License 2.0. This license is permissive and generally compatible with commercial use and linking within closed-source projects.

Limitations & Caveats

Lore is described as "early and evolving quickly," indicating potential for rapid changes and ongoing development. Its primary integration mechanism relies on the MCP protocol, suggesting a focus on specific agent ecosystems.