repowise: open-source codebase intelligence

I'd been building with LLMs since 2023 and owned the AI architecture at my company, so I spent a lot of time watching agents grep the same file four times in a session and still have no idea which parts of the codebase were fragile or why the code was written that way. All of it is in the repo already, just not in a form an agent can use.

So I built Repowise. It indexes your repo once and serves it to your agent over MCP as five layers: dependency graph, git history, docs, architectural decisions, and a code health score from around 25 deterministic markers with no LLM involved. Pulling the context around a single commit costs about 393 tokens instead of roughly 14,000 spent reading files.