repowise vs CodeScene: Code Health, Hotspots and Agent Access

Pick CodeScene if you want a mature, vendor-supported behavioural analysis product with delivery dashboards for managers; pick repowise if you want code health, hotspots and ownership that run on your own machine, cost no licence, and sit in the same index as a wiki, dead code, decisions and agent tools. Both read code and git history to answer the same question: which files are worth fixing first.

CodeScene details are as of October 2026, from its pricing page, product page and MCP server documentation. repowise details are from release 0.56.

repowise vs CodeScene at a glance

CodeScene is a commercial analysis platform with a free tier for open source; repowise is an open-source engine that also generates documentation.

What CodeScene does well

CodeScene made hotspot analysis mainstream: complexity matters most where a file also changes often. Its approach comes from Adam Tornhill's Your Code as a Crime Scene, and it has years of calibration behind it.

• CodeHealth. A single per-file metric, built from code smells, and presented as a trend you can gate on.
• Behavioural analysis. Hotspots, knowledge distribution and bus factor, plus team and delivery insights on the Pro plan.
• Developer loop. An IDE extension, a CLI, pull request integrations and an AI refactoring service (CodeScene ACE).
• Agent safeguards. The CodeScene MCP server runs locally and feeds CodeHealth into assistants such as GitHub Copilot, Cursor and Claude.
• Packaging. Per-author pricing with unlimited login users, an on-premises option, enterprise SSO and a customer success manager.

If a vendor-supported product with delivery metrics for non-technical stakeholders is what you are buying, CodeScene is built for that.

What repowise adds

repowise reads the parse tree, the dependency graph and git history once, and every view comes from that one index. Code health needs no model and no network, and runs inside repowise init and repowise update.

Three health signals instead of one number

Each file gets a defect-risk score (the headline), a maintainability score and a performance-risk score, each from its own findings and weights. A file that is hard to read but rarely breaks is not mixed with one that keeps producing bugs. repowise health lists a Fix first queue, each item with an effort estimate and a verify command.

pip install repowise
repowise init        # scores every file, no LLM key needed
repowise health      # Fix first, KPIs, lowest-scoring files
repowise next        # the short "Do next" list across every layer

Hotspots, ownership and change risk

Hotspots rank files by change frequency and prior bug fixes from full git history, with bus factor, owner and trend per file. repowise risk main..HEAD scores a pending change from its size, spread and the fix history of the files it touches. See git intelligence and change risk.

repowise hotspot ranking showing commits, bug fixes, churn, bus factor and owner per file in the repowise repository

One index for agents

The same data reaches Claude Code, Cursor and other MCP clients through ten tools: get_overview, get_answer, get_context, get_symbol, search_codebase, get_risk, get_change_risk, get_health, get_why and get_dead_code. Beyond health, an agent can ask who owns a file, what a module does, which decision explains a pattern, and what is unreachable.

get_health(targets=["src/api/server.py"])
get_change_risk(revspec="main..HEAD")

A wiki, a graph and dead code from the same pass

repowise also writes a wiki page per file and subsystem, builds a dependency graph, and reports dead code by confidence tier. Browse a live example on FastAPI.

How the health scores compare

We scored the same files with both tools at the same leakage-free commit and checked them against the same defect labels. Under a review budget of 20% of lines, repowise surfaced 2.3x the defects of CodeScene (n = 2,770 · p = 0.003 · 2,770 files scored by both tools at the same leakage-free commit · 2026-08). Paired, same files, same labels. CodeScene is ahead on nominal precision, and its more conservative threshold flags 27 files where we flag 132, which is a real operating-point choice, not a weaker model.

Both tools scored the same 2,770 files at the same leakage-free commit against the same defect labels. Paired tests throughout.

CodeScene's precision lead is not statistically significant, but the behaviour behind it is worth having: it flags about 27 files where we flag 132. That is a more conservative threshold, trading recall for a short list a team will actually work through, and it is a deliberate design choice, not a weaker model. Our AUC edge is marginal, not significant at 0.05.

The method and raw data are on the benchmarks page.

Which one to pick

Pick CodeScene if you need a supported commercial product with delivery and team dashboards, SSO and a customer success contact, and your repositories fit its per-author pricing.

Pick repowise if you want to self-host under AGPL-3.0, keep code and analysis on your own infrastructure, avoid per-seat pricing, or want health, ownership, docs, dead code and decisions in one index your agents can query.

Use both if your managers already rely on CodeScene's delivery insights and your engineers want repowise in the terminal and in their agents. They read the same git history and do not conflict.

Try both on your own repository

pip install repowise
cd your-repo
repowise init --no-prose     # health, hotspots, ownership, wiki: no API key
repowise health
repowise risk main..HEAD

repowise init --no-prose builds the full index without an LLM; add model-written prose later if you want it, from a hosted provider or a local Ollama model. Open the same repository in CodeScene (free trial, or free for open source) and compare the top of each ranking. For agent setup, see how repowise serves context to agents.