Repowise – Local, self‑hosted code‑base intelligence

What it is

Repowise is a Python‑based tool that builds a single local index of a repository’s call‑graph, git history, tests, documentation and design decisions. The index is stored on‑premises (no API key or cloud service required) and can be queried by developers or by any MCP‑compatible coding‑assistant (Claude Code, Cursor, VS Code extensions, etc.).

Core capabilities

Area	What you can ask the index	How Repowise answers it
Code understanding	“What calls this function?” “Where is this feature implemented?” “Why was this module created?”	A cited answer built from the call‑graph (26 languages) and the generated wiki; decisions are linked to ADRs, # WHY: comments, commits and PRs.
Safe change‑management	“What breaks if I modify auth.py?” “Which tests should run for this diff?” “Is this PR risky?”	Symbol‑level blast‑radius from the graph, co‑change history, test‑impact inference and a 0‑10 risk score compared to the repo’s own recent commits.
Code health & debt	“What should we fix first?” “Where is dead code?” “What performance hotspots exist?”	53 deterministic detectors produce a health score per file, ranked fix‑first queues, dead‑code confidence tiers, and refactoring suggestions (extract method/class, split file, etc.).
Documentation	“Give me docs for this module.” “Which docs are out‑of‑date?”	An automatically generated wiki (no LLM needed) plus optional model‑written prose; doc‑drift detection flags markdown that no longer matches the code.
Cross‑repo contracts	“Will this API change break another service?”	Contracts from HTTP, gRPC, OpenAPI, topics are matched across repositories; breaking‑change guards are shown in workspaces.
Security	“Did we just commit a secret?”	Scans for keys, tokens and risky calls in the working tree and full git history, with pre‑commit and CI gates.

How you use it

1. Install – uv tool install repowise (or pipx / pip).
2. Initialize – repowise init --yes --no-prose creates the index (graph, git, health, dead code, docs) without any external API key.
3. Serve – repowise serve runs a local dashboard and an MCP server that coding agents can call.
4. Query – Agents can invoke built‑in tools such as get_overview, risk, distill, or you can use the CLI directly (repowise risk main..HEAD, repowise distill pytest).

A one‑line prompt can also be sent to an MCP‑capable agent:

Read https://docs.repowise.dev/setup.md and set up Repowise in this repository.

The agent will install, index, and wire Repowise automatically.

Evidence & performance

• ‑31.6 % less agent output on a 43‑question benchmark (p < 0.0001).
• 2.3× more defects found than CodeScene (p = 0.003).
• Indexed 58,924 files on a laptop in ≈ 2 h.
• Call‑graph precision 0.976‑0.995 on Go and TypeScript, outperforming six other tools on 7 datasets.
• All analysis runs locally, no LLM calls required for the core layers.

Deployment options

• Self‑hosted – run the dashboard and MCP server on‑premises (AGPL‑3.0).
• Hosted SaaS – repowise.dev offers the same functionality plus the upcoming Foresight feature (models trained on real code‑change patterns) and team workspaces.
• Enterprise – supports multi‑repo workspaces, compliance reporting, sizing guidance, and optional commercial licensing.

Who it’s for

• Developers who want their coding assistants to have instant, accurate context.
• Teams seeking safer pull‑requests, test‑selection intelligence, and risk scoring.
• Organizations that need self‑hosted, zero‑API‑key code‑base intelligence for compliance.
• Anyone looking to prioritise technical debt with data‑driven rankings.

---

All details above are taken directly from the repository’s README; no external claims are added.