Project
Lore
A living knowledge base for your codebase — queryable by humans and AI agents.
The problem
The knowledge that matters most is the knowledge nobody wrote down.
Every mature codebase carries knowledge that isn't written anywhere: why a module is shaped the way it is, which flows actually matter, how errors are meant to be handled, what “the way we do things here” means. New engineers learn it slowly. AI coding agents never learn it at all — they see files, not understanding.
Existing tools cover part of the picture. Vector search is good at fuzzy questions but blind to structure. Code-graph tools know what calls what, but have no idea why. Auto-generated wikis look right on day one and are stale by the end of the month.
What Lore does
Point it at a repository, and it builds the understanding
Four kinds of knowledge, extracted from the code itself and kept current as the code moves.
Architecture
Subsystems, layers, boundaries, and entry points — the shape of the system rather than the file tree.
Flows and use cases
End-to-end paths from an API route, job, or event handler down to the database or an external service.
Code structure
Symbols, dependencies, call graphs, and ownership, with stable IDs so changes can be diffed.
Conventions and decisions
The implicit rules of the repo: patterns, naming, error handling, testing style, architectural decisions.
That knowledge base is exposed through an MCP server, so any MCP-compatible agent — Claude Code, Cursor, your own — can ask precise questions and get grounded answers with references back to the source. The same knowledge base renders documentation and diagrams, so those stay in sync with the code instead of drifting away from it.
The hybrid approach
Three layers, because one index can't answer every question.
A query layer routes each question to the right layer — or combines them — and an incremental indexer updates all three on every commit.
Structural
A code graph extracted by parsing — files, modules, functions, routes, models, and the edges between them.
What, where, and how is it connected?
Semantic
Summaries of every module, flow, and use case, written by a model and embedded for search.
What does this do, and why?
Conventions
Extracted rules and architectural decisions, each linked back to the evidence in the code.
How should new code look here?
Who it's for
Teams with more codebase than institutional memory.
- Engineering teams onboarding new members or maintaining large, long-lived systems.
- Teams adopting AI-augmented development who want coding agents to produce code that looks native to their repo.
- Architects and tech leads who need accurate, current documentation without maintaining it by hand.
Want this pointed at your codebase?
Tell us about the repository and what your team keeps re-explaining to each other, and we'll tell you whether Lore is the right shape for it.