Know what to refactor next.
Stop guessing before planning.
Faultlines turns your git history into a live feature map — with bug hotspots, test coverage per user flow, and the AI-context your agents actually need. One scan. No Jira hygiene.
Free tier · runs locally · code never leaves your machine
Real Faultlines scans across 25 popular open-source repos — grouped by product feature. Reproduce with$pip install 'faultlines[mcp]'
You can't plan a quarter you can't see.
Every sprint, EMs and tech leads burn hours guessing which parts of the codebase are rotting, which user journeys are uncovered, and which “small refactor” is actually a three-week hole. Jira and dashboards don't tell you. The code does — if you can read all of it.
Faultlines reads what your team already wrote — the git history — and turns it into the map you wish you had on day one.
Built for engineering leaders
and the AI agents on their team.
Faultlines turns git history into a feature map, scores each feature on bug density & coverage, and exposes it all through MCP so humans and AI agents work from the same source of truth.
Not a replacement for Sourcegraph.
Not a clone of CodeScene.
Faultlines lives in a specific slice: feature-level, history-based, git-native. Here's where that slice overlaps and doesn't overlap with the tools you probably already pay for.
| Capability | Faultlines | CodeRabbit | Sourcegraph | CodeScene |
|---|---|---|---|---|
| Feature-level clustering with business names | yes(LLM from git history) | no(PR-scoped only) | no(file/symbol search) | partial(manual or heuristic) |
| User-flow detection from codelogin, forgot-password, create-org — named by intent | yes(Haiku 90%+ accuracy) | no | no | no |
| Symbol-line scoped scoringfunction-level blame, not just file-level | yes(line-range blame index) | no | partial(precise code intel) | no(file-level only) |
| Bug-ratio / hotspot per featurechurn weighting, bus factor, age decay | yes(recent bugs weigh 2×) | no | no | yes(10-year leader) |
| Test coverage scoped to a featureaveraged over symbol line ranges, not files | yes | no | no | no |
| PR impact in business language“this PR touches Auth (78% covered) and Billing (45% covered)” | yes(feature- & flow-scoped) | partial(file-level summaries) | no | no |
| MCP-native for AI agentsread-only tools an AI agent calls directly | yes(11 tools) | no | MCP server(beta) | — |
| Runs locally, no source upload | yes(CLI; dashboard sees results only) | SaaS(PR diffs sent) | on-prem tier | on-prem tier |
| Open-source CLI | MIT | no | OSS server(Apache 2.0) | no |
| Total cost — 10-dev teamhow the per-org vs per-seat math actually pencils out | $29/mo(flat per-org) | $120–240/mo($12–24 × 10) | $190–490/mo($19–49 × 10) | $190–360/mo(per author) |
We hold ciphertext and a wrapped key.
Never your code.
A short-lived worker clones, scans, encrypts the feature map with AES-256-GCM, then wipes itself. The master key never leaves AWS KMS — not even for us.
Full security & subprocessorsMap your codebase.
Know what to refactor next.
Scan your first repo in minutes — feature map, risk hotspots, coverage by flow. Free tier · runs locally · OSS.