The open-source CodeScene alternative.
A defect-validated code-health score, measured head-to-head against CodeScene itself on the same 2,770 files, plus the wiki, git intelligence, decisions, and agent-native MCP that CodeScene does not ship. All open source and self-hostable.
CodeScene proved that behavioral code analysis works, and it is the tool repowise's own benchmark is measured against. The question is whether you want that analysis as a closed, per-author analytics silo, or as an open layer that also documents, explains, and feeds your AI agents.
repowise takes the same behavioral foundation, hotspots, ownership, coupling, and a defect-validated health score, and makes it open, reproducible on your own repo, and part of a wider intelligence layer rather than a standalone dashboard. The comparison is not hypothetical: repowise and CodeScene were scored on the exact same corpus.
Which one is right for you?
Choose repowise if
- You want code health that is open source and self-hostable, with every heuristic inspectable
- You want a benchmark measured directly against CodeScene, not just a citation
- You want health plus an auto-generated wiki, architectural decisions, and agent-native MCP in one tool
- You want dead-code detection alongside health, with no separate purchase
- You prefer per-repo and per-seat pricing over a per-active-author model
Choose CodeScene if
- You need AI auto-refactoring with a validation model (CodeScene ACE)
- You need DORA, delivery, and portfolio management dashboards
- You want decades of peer-reviewed behavioral-code-analysis research behind the score
- You need mature support for 30+ languages today
repowise vs CodeScene
| Capability | repowise | CodeScene |
|---|---|---|
| Deterministic code-health score | Included | Included |
| Open source and self-hostable | Included | Not included |
| No API key needed for the first index | Included | Not included |
| Defect validation, measured head-to-head on the same corpusCodeScene cites peer-reviewed research; repowise ships a direct benchmark against CodeScene you can rerun | Included | Partial support |
| Hotspots, ownership, and bus factor | Included | Included |
| Change coupling / hidden coupling | Included | Included |
| Three co-equal health signals (defect, maintainability, performance) | Included | Partial support |
| Auto-generated wiki and documentation | Included | Not included |
| Architectural decision records | Included | Not included |
| Multi-repo workspaces | Included | Partial support |
| VS Code extension | Included | Not included |
| Agent-native MCP context (overview, answers, risk, why)CodeScene's MCP is focused on health insights | Included | Partial support |
| AI code provenance (agent attribution) | Included | Not included |
| Dead code detection | Included | Not included |
| AI auto-refactoringCodeScene ACE | Not included | Included |
| DORA, delivery, and portfolio dashboards | Partial support | Included |
| 30+ language supportrepowise: 16 languages, full-tier for 11 | Partial support | Included |
Self-assessed against publicly documented features as of June 2026. A dash means partial or limited support. Vendor capabilities change, so please verify against CodeScene's current docs before deciding.
Measured against CodeScene, not just compared to it.
This is the one page where repowise's own benchmark result is a direct answer to the question 'how does this stack up against CodeScene', because that is literally what was measured.
Same corpus, same labels
2,770 files across 9 languages from 21 repos, scored by both tools against the same real defect labels. Neither tool got a corpus tuned to its own strengths.
Cross-project, leakage-free
A file's own future commits never inform its own score. This is a forward-looking prediction test, not an after-the-fact association on data the model already saw.
ROC AUC 0.74
95% CI 0.68 to 0.79, up to 0.90 per repo. This is the probability that repowise ranks a randomly chosen buggy file above a randomly chosen clean one, measured on held-out repos.
2.3x more defects caught
Under a fixed review budget (the same number of files a team would actually have time to review), repowise's ranking surfaces more of the files that really did ship bugs than CodeScene's ranking does, on that same corpus.
repowise also publishes a separate, simpler self-check on any repo you index: 16/20 lowest-health files had a bug fix in the last 6 months, 3.3x the 24% baseline (80% vs 24%).That number is an association on the repo's own history, not the leakage-free forward test above, and the two are reported separately so neither one is quietly doing the other's job.
Our own repo, public and free.
Every claim on this page is checkable on a real, live index: repowise's own OSS repository, indexed by repowise itself and open to anyone.
Auto-generated architecture, modules, and entry points, something CodeScene does not attempt.
Every file scored 1 to 10 across defect, maintainability, and performance, deterministic and free.
Unreachable files and unused exports, tiered by confidence, with no equivalent on CodeScene.
Health is the start, not the whole story.
The same behavioral signals CodeScene pioneered, plus the context layer it was never built to provide.
A score you can audit, not just trust
Every marker and weight is open source. The defect-validation benchmark runs on your own repo, so you can confirm the score finds your bugs rather than taking a vendor's word or a citation for it.
- Cross-project ROC AUC 0.74, benchmarked directly against CodeScene
- 2.3x more defects under a fixed review budget vs CodeScene
- On a typical repo, the self-check shows 16 of the 20 worst files had a recent bug fix, 3.3x the baseline
- AGPL-3.0 (OSS 0.35.0): inspect, fork, self-host
Health, docs, decisions, and agent context together
CodeScene is a code-analytics product. repowise puts health alongside an auto-generated wiki, architectural decision archaeology, git intelligence, agent provenance, and ten MCP tools, so the same index serves your quality goals and your AI agents.
- Auto-generated wiki, rebuilt on every commit
- Architectural decisions mined from eight sources, including git archaeology and agent transcripts
- Multi-repo workspaces for cross-service understanding
- MCP tools for Claude Code, Cursor, Cline, and Codex, plus a VS Code extension
No per-author tax, and free if you self-host
repowise is free and open source to self-host, with hosted tiers priced per repo and per seat. CodeScene is free only for open-source projects and otherwise charges per active author.
- Self-host the full platform at no cost under AGPL-3.0
- No API key needed for the first index; bring your own LLM key only for prose
- Zero telemetry, code never leaves your infrastructure
- Commercial license available when you need it
A ranking you can rerun, not a one-off report.
Deterministic markers, not LLM judgment
Both repowise's code health and CodeScene's CodeHealth are computed from deterministic, rule-based markers rather than an LLM call, which is why the head-to-head benchmark above is a fair, reproducible comparison and not two different kinds of scoring being forced together.
Same score, every run
21 markers score every file the same way each time on repowise's side. No temperature, no prompt drift, no different answer on a re-run of an unchanged file.
The prose layer is separate and optional
repowise's wiki narrative is the one LLM-touched part of the pipeline, and it sits on top of the deterministic health, dead-code, and git-intelligence layers rather than replacing them. Skip it and the score, ranking, and benchmark are unaffected.
Validated against real bugs, on CodeScene's own comparison ground
Cross-project ROC AUC 0.74 (95% CI 0.68 to 0.79, up to 0.90 per repo) against real defect labels, measured on the same corpus CodeScene was scored on, not a similarity score between a prompt and a diff.
The honest version
CodeScene is a mature, respected product, and there are places it leads today. Its CodeScene ACE offers AI auto-refactoring with a validation model, which repowise does not. It ships DORA, delivery, and portfolio management dashboards for engineering organizations. Its CodeHealth metric is backed by years of peer-reviewed research, including an award-winning study, and it supports 30+ languages. If those are your priority, CodeScene is a strong choice. repowise wins when you want openness, a benchmark measured directly against CodeScene rather than cited from a paper, a full context layer, and agent-native access, at team-friendly pricing.
What repowise costs.
The repowise core is open source under AGPL-3.0 and free to self-host, with every heuristic public. Hosted tiers: Free for public repos, Pro at $15 per month, Teams at $60 per month, and custom enterprise licensing when you need it. CodeScene's pricing changes, so verify it on their site, then compare it with a tier you can read line by line.
Questions, answered
Is repowise a good CodeScene alternative?
Yes, if you want behavioral code-health analysis that is open source and self-hostable, and you want more than health alone. repowise scores code health from 21 deterministic markers, validated against real defects head-to-head against CodeScene itself, and bundles it with an auto-generated wiki, git intelligence, architectural decisions, agent provenance, and ten MCP tools for AI agents. CodeScene remains the better fit if you specifically need its AI auto-refactoring or its DORA and delivery dashboards.
Is repowise open source? CodeScene is not.
Yes. The repowise core is open source under AGPL-3.0, so every marker, weight, and scoring rule is public and inspectable, and you can self-host the whole platform with one pip install and no API key for the first index. CodeScene is closed source and hosted or on-prem under a commercial license.
Does repowise have a code-health score like CodeScene's CodeHealth?
Yes. repowise scores every file 1 to 10 from 21 deterministic markers (complexity, nesting, cohesion, clones, change entropy, co-change scatter, ownership dispersion, prior-defect history, and more), with no LLM, in under 30 seconds on a 3,000-file repo.
How does repowise's ROC AUC 0.74 number against CodeScene actually work?
This is a real head-to-head benchmark, not a marketing estimate: repowise and CodeScene were both scored against the same 2,770 files across 9 languages from 21 repos, using the same defect labels, cross-project and leakage-free (no file's own future commits inform its own score). repowise reached cross-project ROC AUC 0.74 (95% CI 0.68 to 0.79, up to 0.90 per repo), catching 2.3x more defects than CodeScene under a fixed review budget on that same corpus. CodeScene's own published research is a genuine strength of theirs, and this benchmark is repowise's answer measured on the same footing.
Does repowise do AI auto-refactoring like CodeScene ACE?
No. CodeScene ACE offers AI auto-refactoring with a validation model, and repowise does not. repowise focuses on understanding, risk, and context: finding the riskiest code and giving your agents the context to fix it well.
Does repowise detect dead code? CodeScene does not have an equivalent.
Yes. repowise's dead-code detector finds unreachable files, unused exports, and zombie packages, tiered by confidence, with zero LLM calls. This runs alongside the health score and git intelligence, not as a separate purchase. CodeScene does not ship an equivalent dead-code detector.
Is code health one blended score, or three separate signals?
Three separate, co-equal signals: defect risk (the primary 1 to 10 score, the one validated against CodeScene on the same corpus), maintainability (readability and change-cost smells), and performance (static I/O-in-loop and N+1 shapes). They are never blended into a single number, which keeps the defect score's validation claim clean, since maintainability noise never dilutes it.
Is repowise cheaper than CodeScene?
For most teams, yes. repowise is free and self-hostable under AGPL-3.0, and the hosted tiers are priced per repo and per seat rather than per active author. CodeScene is free only for open-source projects and otherwise prices per active author.
Can repowise give AI coding agents codebase context?
Yes, and this is a core difference. repowise exposes the whole index through ten MCP tools (get_overview, get_answer, get_context, get_risk, get_why, and more) so Claude Code, Cursor, Cline, and Codex answer from a real model of your code. CodeScene's MCP surface is narrower and focused on health insights.
The defect-validated score, in depth.
21 markers, the ROC AUC methodology, and how to read a worklist.
Hotspots, ownership, coupling, bus factor.
Health, ownership, and AI-debt for leaders.
The dedicated roundup this page distills.
The full head-to-head writeup.
The validation study behind the ROC AUC numbers.
See the wiki and code health on a real, public repo.