Most coding agents repeatedly rediscover the same repository context — grep, open files, re-read history — burning model calls and tokens. Repowise flips that cost: compute the graph, git signals, generated wiki, decisions and deterministic code-health once on your machine, and let agents and humans query that index with a small set of task-shaped MCP tools.
What Sets It Apart
- Deterministic, evidence-first index: the index builds a multi-layer artifact (call graph, git hotspot/ownership, generated docs, mined decisions, and 1–10 file health scores) with zero LLM calls. So what: reviews, CI gates and agent hooks run reproducibly and without sending code to an external model by default.
- Task-shaped MCP surface for agents: ten concise tools (get_overview, get_context, get_risk, get_health, get_dead_code, etc.) return complete, cited context in one round-trip. So what: agents make far fewer calls and produce less noisy output while keeping the reasoning traceable to files and commits.
- Graph-aware code health and refactoring plans: 50+ deterministic detectors across defect risk, maintainability and performance, plus concrete, graph-aware refactor plans (Extract Method/Class, Break Cycle, Move Method). So what: teams get a ranked, actionable backlog rather than vague warnings.
- Cross-repo workspace intelligence and CI gates: HTTP/OpenAPI/proto contract mapping, cross-repo blast radius, and change-risk scoring for PRs. So what: large estates can detect breaking consumer impact and run only the tests a change actually exercises.
Who it's for and tradeoffs
Great fit if you run or integrate coding agents, need deterministic, reproducible pre-merge analysis, or maintain multi-repo estates where change impact crosses service boundaries. It is designed for teams that want the analysis to stay inside their network and prefer evidence-backed answers to probabilistic LLM reviews.
Look elsewhere if you only need a lightweight call-graph extractor (other tools build a graph faster), or if you require a fully managed SaaS where your source must be uploaded (hosted options exist but the core value is self-hosted evidence). Enabling model-written prose or optional generation features sends prompts to whatever provider you configure; the deterministic layers remain offline by default.
Where it fits
Use it as the context layer for coding agents (Claude Code, Codex, Cursor, VS Code extension), as a CI gate for change-risk and patch-coverage, and as an engineering dashboard to prioritize refactors and remove dead code. The index output lives locally under .repowise or in your self-hosted backing store when run at scale.
