Every engineer who has joined a new team knows the feeling: a repository with years of history, half-documented conventions, and a dozen implicit assumptions baked into the code. Onboarding docs go stale. Comments lie. The only source of truth is the code itself — and reading all of it doesn't scale.
CodeMind AI exists to close that gap. Instead of a chatbot that's read about your code, it's a system that actually indexes your repository: parsing every file with tree-sitter, extracting symbols and import relationships, chunking content for retrieval, and building a real dependency graph. When you ask a question, the answer comes with citations — exact file paths and line ranges — because "trust me" isn't good enough when you're about to ship a fix.
We started narrow on purpose. The first version does one thing well: index a repository and answer questions about it with real citations, backed by a tree-sitter-derived symbol graph rather than a vector-only guess. Everything since — the architecture graph, bug and security scanning, PR review, fix proposals — has built on that same foundation of "don't claim something the system can't actually verify against the code."
That's still the principle behind every feature we ship: if CodeMind AI tells you something about your codebase, it should be able to point at exactly where that came from.