{ "markdown": "\n\n
\n\n\"repowise:\n\n

Understand your codebase without paying your agent to rediscover it.

\n\n

Repowise indexes your code, dependency graph, git history, tests,
\ndocumentation, and decisions once, then gives agents and developers cited answers,
\nchange impact, and concrete code-health fixes.

\n\n

\n \"Open\n

\n\n\"Repowise\n\n\n\n\n\n\n\n\n\n\n\n\n

−31.6%

97.2%

2.3×

less agent output
3.8 vs 7.2 tool calls
n=43 · p<0.0001		smaller context payload
393 vs 13,984 tokens
30 Flask commits		more defects surfaced
same 20%-of-lines budget
2,770 files · p=0.003
\n\n

Graph accuracy leader at matched coverage.
\nNo tool finding as much was more precise in all 7 compiler-graded cells.
\n5 tools · 37,853 oracle edges

\n\n

Zero LLM calls for graph, risk, health, tests,\ndead code, and PR review. Generated prose is optional. Every benchmark publishes its\nsample, method, limitations, and losing rows.

\n\n

\nFree and self-hosted · core analysis stays on your infrastructure · no API key needed ·\nAGPL-3.0 or commercial\n

\n\n

\n \"Explore\n \"Repowise\n \"PyPI\n \"License:\n

\n\n

\n Why Repowise ·\n Agents ·\n Changes ·\n Code health ·\n Workspaces ·\n Evidence ·\n Enterprise ·\n Docs\n

\n\n
\n\n---\n\n\n\n## One index. Three ways to use it.\n\n| **Understand the code** | **Change it safely** | **Improve it continuously** |\n|---|---|---|\n| Ask cited questions · explore architecture and execution flows · read always-current docs · recover the decisions behind the code | See symbol-level blast radius · run only the tests a diff exercises · catch missing companion files · detect breaking contracts before merge | Find defect-prone files · separate maintainability from performance risk · remove dead code · hand concrete, graph-aware refactoring plans to an agent |\n\nThese are not disconnected scanners. The graph locates what git history flags; code\nhealth measures it; tests show what guards it; decisions explain why it exists; and\nthe same evidence reaches your agent, editor, pull request, local dashboard, and\ncross-repository system map.\n\n
\n\"The\n

A dashboard tour recorded on this repository. The same local index powers the UI,\nMCP tools, editor views, and PR analysis. No API key and nothing uploaded.

\n
\n\n### Pick your front door\n\n| If you care about… | Start here |\n|---|---|\n| **A coding agent that understands the repository** | Repowise finds the right files, returns task-shaped context in fewer calls, and proactively supplies decisions and risk. [For agents ↓](#your-agent-stops-guessing) |\n| **Safer pull requests and faster test feedback** | Get change risk, symbol-level callers, co-change partners, and a measured or graph-inferred test run list before merge. [Change intelligence ↓](#know-whats-dangerous-before-you-merge) |\n| **Finding and fixing the code most likely to hurt you** | A defect-validated 1–10 health score across defect risk, maintainability, and performance, followed by the concrete refactoring plan. [Code health ↓](#code-health) |\n| **Understanding an estate, not one repository** | Match backend and frontend contracts, catch breaking providers, map downstream services, enforce architecture rules, and query every repo through one MCP endpoint. [Workspaces ↓](#past-one-repo) |\n| **Rolling this out across an engineering organization** | Keep analysis on your infrastructure, give agents and reviewers the same evidence, and add commercial licensing, security controls, custom extensions, and SLA-backed support. [Teams and enterprise ↓](#for-teams-and-enterprises) |\n\n\n\n## Start in minutes (no API key)\n\n```bash\npip install repowise\ncd /path/to/your/repo\nrepowise init --no-prose -y\nrepowise serve\n```\n\nThat builds the graph, git, decisions, health, dead-code and structural-wiki layers\nlocally. Connect Claude Code, Codex, Cursor or any MCP host, or open the dashboard.\n`init` wires Claude Code automatically. Then ask your agent: *\"Use Repowise\n`get_overview` to summarize this repository\"* or *\"What breaks if I change\n`src/auth.py`?\"*\n\n[Full setup, every agent, and optional model-written prose →](docs/start/QUICKSTART.md)\n\n---\n\n## Your agent stops guessing\n\nEvery question your agent asks about a repository has an answer that could have been\ncomputed ahead of time. *Who calls this function? What breaks if I change it? Why is\nit written this way? Which files are actually dangerous?* Without an index, the agent\nrediscovers that answer on every task: grep, read, re-read, forget.\n\nRepowise exposes **ten task-shaped MCP tools** to Claude Code, Codex, Cursor, VS Code\nand anything else that speaks MCP: graph, git, docs, decisions, and ten MCP tools\nbehind one index. See [the canonical surface](#the-ten-mcp-tools). Most tools are built around data entities (one\nfile, one symbol), which forces agents into long chains of sequential calls. These are\nbuilt around **tasks**: pass several targets in one call, get complete context back.\n\nBecause the exploration work is already done, that phase mostly disappears. In a\nmeasured agent loop across 43 questions on `django/django`, Repowise cut the agent's\nown output by **31.6%** (p<0.0001) and reached the answer in **3.8 tool calls\ninstead of 7.2**. That is the end-to-end result.\n\nOne mechanism is much larger but narrower: loading a commit's context through\n`get_context` costs **393 tokens instead of 13,984**, or 97.2% less. That is one\nretrieval payload, not a claim of 97.2% total agent savings. Both measurements and\nevery competitor row are published in [the benchmark report](docs/BENCHMARKS.md).\n\n**And it arrives without being asked.** Optional [hooks](docs/agent/HOOKS.md) push\ncontext into the session at the moment it matters: the governing architectural\ndecision when your agent edits a file that decision covers, a warning when it touches\na file with a run of recent bug fixes, a compact briefing at session start. Repowise\nalso generates your `CLAUDE.md` and `AGENTS.md` from the real index, so even an agent\nwith no MCP support starts informed.\n\n**It learns from how you actually work.** Repowise reads your own agent transcripts\nfor the corrections you keep making (\"use the shared HTTP client, not raw requests\")\nand turns the durable ones into tracked decisions it delivers back later. The wiki\ngeneration budget tilts toward the modules you and your agent ask about most. All\nlocal, all deterministic, no extra LLM calls.\n\n
\nWhat the index builds\n\n| Foundation | What it contributes |\n|---|---|\n| **Graph** | File + symbol dependencies across 25 AST-parsed languages, confidence-stamped call resolution, communities, centrality, cycles, and execution flows |\n| **Git** | Hotspots, ownership, co-change, bus factor, and bug-fix history: behavioral signals static analysis cannot see |\n| **Docs** | A wiki for every module and file, rebuilt incrementally with freshness and confidence scoring plus hybrid search |\n| **Decisions** | Architectural rationale mined from five index-time sources plus human and agent capture, each claim traced to evidence |\n| **Code health** | 49 deterministic detectors across defect risk, maintainability, and performance, followed by concrete refactoring plans |\n\nThe structural wiki needs no model. Model-written prose is an optional upgrade, one\npage or directory at a time. Six of the seven decision sources are deterministic too;\nonly comment archaeology needs a provider.\n\n[The intelligence layers →](docs/layers/INTELLIGENCE_LAYERS.md) ·\n[How the graph earns trust →](docs/layers/GRAPH.md)\n\n
\n\n### Also: stop paying for output nobody reads\n\nMost of what an agent reads back from a shell command is noise: 300 lines of passing\ntests wrapped around 4 failures, full commit bodies when it asked \"what changed\nrecently\". `repowise distill ` compresses command output **before the agent reads\nit**, errors first, exit code preserved.\n\n```bash\nrepowise distill pytest # 61% fewer tokens, all 11 failure lines kept\nrepowise distill git log -50 # 89% fewer tokens\nrepowise saved # what distillation saved you, in tokens and dollars\n```\n\nNothing is lost. Every omission leaves an inline `[repowise#]` marker that\n`repowise expand ` reverses in full, so the agent can always pull the detail back\nwithout re-running the command. Small outputs pass through untouched. An opt-in hook\nrewrites noisy commands automatically, shown to you for approval first.\n\n
\n\"repowise\n

The Costs dashboard tallies both savings surfaces, priced at your own agent's model. Example from a week of heavy local use.

\n
\n\nFull guide: **[docs/agent/DISTILL.md →](docs/agent/DISTILL.md)**\n\n---\n\n## Know what's dangerous before you merge\n\nFour deterministic signals, all computed from the graph and git history, no LLM:\n\n- **Change risk.** Score any commit or `base..HEAD` range **0-10** from the shape of\n the diff, ranked against your repo's own recent commits. PR mode returns directives\n rather than vibes: `may_break`, `missing_cochanges`, `missing_tests`, `tests_to_run`.\n One command: `repowise risk main..HEAD`. ([reference →](docs/layers/CHANGE_RISK.md))\n- **Bug history.** Which files and symbols actually get bug-fixed, and how recently.\n Doc, test and config commits are filtered out so the count means what it says, and a\n file with a run of recent fixes gets flagged as a bug magnet while you edit it.\n ([reference →](docs/layers/BUG_HISTORY.md))\n- **[Test intelligence](#which-tests-cover-this-file-without-a-coverage-report).** Which\n tests reach a file and which ones a diff actually exercises, from the call graph,\n with or without a coverage report.\n ([reference →](docs/layers/TEST_INTELLIGENCE.md))\n- **Change coordination.** Which other open branches edit the files you are editing,\n every row saying why it is listed (`same file`, or a co-change pair with the commit\n counts behind it), and whether the diff in front of you is one change or several\n groups the index links nothing between. Both stay quiet when there is nothing to\n report. `repowise overlap` and `repowise risk`.\n ([reference →](docs/layers/CHANGE_RISK.md#branch-overlap))\n\nPlus the free **[Repowise PR Bot](#the-pr-bot)**, which puts all of it on every pull\nrequest. Zero LLM calls.\n\n---\n\n## Which tests cover this file, without a coverage report\n\nIngest LCOV, Cobertura or Clover and you get the measured answer. **Most\nrepositories never produce one**, so the graph answers instead: a test file that\nimports a source file *reaches* it, which is a recorded edge rather than the\nname-shaped guess everything else falls back to.\n\nThat fallback fails in both directions, and this repo is the proof. Five of its\nsix worst bug-magnet files have no test named for them and read as untested while\nthe graph names 3 to 23 test files each. The sixth is worse: matching on basename\npaired the *health* engine with the *distill* engine's tests and called it tested.\n\n```bash\nrepowise impacted-tests main..HEAD # only the tests this diff actually exercises\nrepowise health # untested hotspots, now graph-aware\n```\n\nDogfooded against a real coverage run --contexts=test:\n95.7% precision on what reaches a file and 97.5% on the\nrun list, at a 100% hit rate, against 72.1% and 94.8% for the one-hop\nimport walk this replaced. The two tiers are never averaged: rows are stamped\nbasis: \"measured\" or \"inferred\", measured wins outright\nwhere both can answer, and the inferred tier may never produce a percentage.\nSound as a floor, unsound as a quantity, and labelled so.\nTest intelligence →\n\n---\n\n## The PR bot\n\nInstall the [GitHub App](https://github.com/apps/repowise-bot) and the index shows up\nwhere the decision actually gets made. One comment per pull request, edited in place on\nevery push rather than reposted, and **a green PR gets no comment at all**.\n\nSee a real comment on a real PR, not a mockup:\n[repowise-dev/repowise#1204](https://github.com/repowise-dev/repowise/pull/1204).\n\nWhat decides a review is inline. What is context sits behind one fold, so the comment\nstays about seventeen rows whatever it finds.\n\n- **Blast radius, at symbol level.** The contracts this PR changed and every caller of\n them in a file the PR does not touch. Importing a module says nothing about whether\n the function you changed is the one being called, so file-level impact is the wrong\n altitude for the question a reviewer actually has.\n- **Before you merge.** The tests that import your changed files, and the files that\n changed alongside them in past commits but are missing here.\n- **A Check Run that can gate the merge**, with annotations on the specific lines the\n PR added. Advisory by default.\n- **Change risk**, scored against the repository's own commit distribution rather than\n an absolute scale, so it stays meaningful on a repo whose typical commit is large.\n- **AI vs human authorship** of the changed files, with the average health of each.\n- Then hotspots, hidden coupling, declining health, dead code and the change map, one\n fold down.\n\n### And a page the comment links to\n\nMarkdown runs out. The comment shows three callers and says \"+6 more\"; the page shows\nall nine. Public, no sign-in, on a repository the reader has never seen.\n\n\"The\n\nThe page leads with change risk and newly introduced findings, then maps every\nchanged file and outside caller across the repository.\n[See it live →](https://repowise.dev/pr/repowise-dev/repowise/1204)\n\n**[Install the PR bot →](https://github.com/apps/repowise-bot)** ·\n[how it works →](https://www.repowise.dev/bot)\n\n---\n\n\n\n## ★ Know exactly what to fix\n\nA score that says *\"this file is risky\"* is where most tools stop. Repowise scores\nevery file, locates where the risk concentrates, and then names the specific fix.\n\n
\n\"repowise\n
\n\nEvery file is scored 1-10 by **49 deterministic detectors** (McCabe complexity, brain\nmethods, LCOM4 cohesion, god classes, native Rabin-Karp clone detection, untested\nhotspots, change entropy, prior-defect history and more), split into three lenses:\n**defect risk**, **maintainability**, and **performance**: static N+1 and I/O-in-loop\nrisk traced *across* files through the call graph, where file-local linters found **0**\nof the cross-function cases and repowise surfaced ~90. Only **26** of the 49 are\npermitted to move the defect number, because that is the number carrying published\naccuracy claims.\n\n> **Zero LLM calls, zero cloud, zero new runtime dependencies.** Pure Python over\n> tree-sitter and git data, **under 30 seconds** on a 3,000-file repo, a budget\n> enforced by a CI test, not an estimate. Marker weights are **calibrated against a\n> real defect corpus, not hand-tuned**: every file scored at a commit preceding the\n> bug window so nothing leaks backward, and an L2-logistic fit with file size as an\n> explicit control, so a marker only earns weight for defect lift *beyond* being big.\n> Only the learned constants ship.\n\n**It proves itself on your repo, not just on a benchmark.** After every index,\nRepowise checks its own flags against your git history and reports what it found:\n*\"16 of the 20 lowest-health files had a bug fix in the last 6 months, 3.3x the 24%\nbaseline.\"* If that number is bad on your codebase, you will see it. (It is an\nassociation on your indexed history, not a forward prediction, the leakage-free\nversion is [in the benchmarks](docs/BENCHMARKS.md#5-code-health-predicts-defects).)\n\nThen it names the fix. Not \"this class is too big\", but **Extract Class**, **Extract\nHelper**, **Move Method**, **Break Cycle**, **Split File**, or **Extract Method**, with\nthe exact methods, edges and symbols that move, the **blast radius** of callers and\nco-changing files that have to move with them, and a graph-aware ranking so a fix on a\ncentral hub outranks the same fix on a leaf. Extract Method goes down to an\nintra-procedural dataflow pass that lifts the exact span and infers a\nbehavior-preserving signature.\n\n```bash\nrepowise health # KPIs and lowest-scoring files\nrepowise health --refactoring-targets # ranked, concrete plans\nrepowise health --trend # snapshots plus declining-health alerts\n```\n\nThe dashboard renders each plan as a card with a copy-to-agent button. An optional LLM\nstep, never in the indexing path and only on request, expands any plan into generated\ncode and a unified diff.\n\nValidated on 21 open-source repos across 9 languages (2,826 files,\nscored at a fixed point and checked against the following 6 months of bug fixes,\nkeyword-labelled): ROC AUC 0.737 [0.683, 0.787]. The signal is\ncorrelated with file size and weakens sharply within a fixed size band, which we report\nrather than bury. Independently recomputed from the raw data.\n\nAgainst CodeScene, the leading commercial code-health tool, on the\nsame 2,770 files and the same defect labels, ranking by repowise health surfaces\n2.3x the defects under a fixed review budget (paired, p = 0.003).\nFull head-to-head, methodology and limitations →\n\nGuides: **[code health](docs/layers/CODE_HEALTH.md)** · **[refactoring](docs/layers/REFACTORING.md)**\n\n---\n\n## See all of it\n\n`repowise serve` starts the full web dashboard next to the MCP server. No separate\nsetup, all local.\n\n\n\n\n\n\n\n\n\n\n
\"Architecture
Architecture · the dependency graph, laid out and explorable, with per-node context and change coupling		\"Code
Code Health · every file as a bubble, hover any one to inspect its score, size, coverage and findings
\"Chat
Chat · ask the codebase a question, answers cite the files and pages they came from		\"Docs
Docs · auto-generated wiki pages for the whole codebase, with confidence and freshness badges
\n\nAlso in there: **Chat** (ask the codebase in natural language) · **Docs** (the\ngenerated wiki, with Mermaid and a graph sidebar) · **Architecture** and **C4**\n(Context → Containers → Components) · **Knowledge Graph** plus a zoomable canvas map ·\n**Risk**, **Hotspots**, **Coupling** and **Blast radius** · **Contributors** ·\n**Decisions** (evidence drawer and evolution timeline) · **Symbols** · **Security** ·\n**Dead code** · **Stats** · **Costs** · **Workspace**.\n\nEvery view and what each one answers: **[docs/start/DASHBOARD.md →](docs/start/DASHBOARD.md)**\n\n---\n\n\n\n## One intelligence layer across your software estate\n\nReal systems are not one repository, and the expensive failures live in the gaps\nbetween them. Change a backend contract and Repowise can name the frontend calls that\nconsume it, the services downstream, the historical companion files missing from the\nchange, and the architecture rule the new dependency violates before it ships.\n\n| Workspace intelligence | What it answers |\n|---|---|\n| **Contract map** | Which services provide and consume each HTTP, gRPC, event, socket, and data contract? Links retain exact/candidate confidence and the source evidence. |\n| **Cross-repo blast radius** | If this provider changes, which downstream services **will break** through structural dependencies, and which ones **may drift** through historical co-change? |\n| **Breaking-change guard** | Was an endpoint removed or a typed contract changed incompatibly, and which exact consumer files call it? |\n| **Test impact** | Which tests in the consumer repos should run for this provider change, measured from coverage or inferred from the call graph, and which links could not be determined? |\n| **Architecture as code** | Does the live system graph violate declared dependency rules or contain cycles? `repowise workspace check` gates CI. |\n| **Architecture health** | How coupled is the estate? Track propagation cost, the cyclic core, service roles, and a deterministic 1–10 architecture score. |\n| **Federated context** | One dashboard and one MCP server answer across every repository while preserving repo-level evidence. |\n\nThe system map models **services**, not merely repository boxes, and never conflates a\nreal contract with “these files often changed together.” Field-level breaking diffs\ncurrently require a gRPC schema; HTTP supports endpoint-level removal detection.\n\n**[Workspace guide and exact support matrix →](docs/scale/WORKSPACES.md)**\n\nWorktrees and updates stay lightweight: a linked worktree seeds its index from the base\ncheckout automatically, and post-commit hooks, file watching, webhooks, or polling keep\neach repository and the cross-repo graph current.\n\n---\n\n## In your editor\n\nThe **Repowise** VS Code extension puts the index where code actually gets written:\nknow what your change breaks before you push (riskiest files ranked, what is\ndownstream, forgotten companion files, missing tests, suggested reviewers), health in\nthe gutter and status bar, callers and ownership on hover, refactoring plans as\nCodeLens, and the full dashboards inside the editor. One install also registers the MCP\nserver with VS Code, so the same local index serves both you and your agent, and\nexposes six tools to GitHub Copilot. Quiet by default, everything toggleable, nothing\nleaves your machine.\n\nInstall from the Marketplace (search **Repowise**) or Open VSX, then run **Repowise:\nSet Up This Repository**. Guide: **[docs/agent/VSCODE.md →](docs/agent/VSCODE.md)**\n\n---\n\n## Supported agents\n\n**Six agents wired end to end · two at the Full tier · every other MCP host one\npaste away.**\n\n
\nSee integration tiers and supported agents\n\n

\n Full tier  \n \"Claude\n \"Codex\n

\n

\n Good tier  \n \"VS\n \"Cursor\"\n \"OpenCode\"\n \"Hermes\"\n

\n\n**Full** is every surface repowise has: MCP tools, skills, slash commands, a managed\ninstructions file, hook-level interception of tool calls, and transcript mining after\nthe session. **Good** is the honest half of that: MCP tools and the config to reach\nthem, but no hook-level interception and no transcript mining. A Good-tier agent can\nask repowise anything; repowise never sees the tool calls in between. The tier is\ncomputed from what each integration actually wires, so this list cannot claim a depth\nthe code does not have.\n\nEverything else that speaks MCP is one snippet away. `repowise agents print-config\nclaude-code` prints a stdio server entry to paste into Cline, Windsurf, Zed, Gemini\nCLI or any other host that keys on `mcpServers`, and repowise writes nothing.\n\nAdding an agent takes **one descriptor file and one registry line**, with no changes to\nthe orchestrators. Full matrix and the contributor recipe:\n**[docs/agent/INTEGRATIONS.md →](docs/agent/INTEGRATIONS.md)**\n\n
\n\n---\n\n## Supported languages\n\n**25 languages parsed to AST · 39 on a five-rung ladder · framework-aware across\nall of them.**\n\n\"Do you support X\" has five useful answers, not two, so languages land on a\nladder and every rung says what it buys you.\n\n
\nSee the complete language ladder\n\n

\n Full tier  \n \"Python\"\n \"TypeScript\"\n \"JavaScript\"\n \"Svelte\"\n \"Vue\"\n \"Java\"\n \"Kotlin\"\n \"Go\"\n \"Rust\"\n \"C++\"\n \"C#\"\n \"Scala\"\n \"Ruby\"\n

\n

\n Good tier  \n \"C\"\n \"Swift\"\n \"PHP\"\n \"Dart\"\n \"Object\n \"GDScript\n \"VB.NET\"\n \"Elixir\"\n \"F#\"\n \"Objective-C\"\n  · Partial  \n \"Luau\"\n \"Razor\n

\n\nBelow those two rungs the ladder keeps going, and a language on a lower rung is\nstill doing real work rather than being ignored:\n\n| Rung | Languages | What you get |\n|---|---|---|\n| **Full** (13) | Python · TypeScript · JavaScript · Svelte · Vue · Java · Kotlin · Go · Rust · C++ · C# · Scala · Ruby | The whole pipeline: AST symbols, import resolution, a resolved call graph, heritage, docstrings, framework edges, **and code-health markers** |\n| **Good** (10) | C · Swift · PHP · Dart · Object Pascal · GDScript · VB.NET · Elixir · F# · Objective-C | All of the above except the full health suite |\n| **Partial** (2) | Luau / Roblox · Razor / Blazor | Luau: AST symbols and `require()` resolution, Rojo and `.luaurc` aware. Razor: component symbols, `@code` and component-tag call edges, C# health markers; no import resolution yet |\n| | | ⎯⎯ *tree-sitter parsing stops here; the rungs below come from git and imports* ⎯⎯ |\n| **Lightweight** (6) | Clojure · Haskell · Lean 4 · Erlang · HTML · QML | A real file-to-file import graph, and no symbol-level claims |\n| **Structural** (8) | R · Zig · Julia · Elm · OCaml · Crystal · Nim · D | Git history: blame, hotspots, co-change, ownership, bug history |\n\n**Every language ships in the open-source distribution.** None is gated behind\nthe commercial licence, and none will be. Languages on the way up the ladder,\nincluding **COBOL**, are on the\n**[roadmap →](ROADMAP.md#languages)**.\n\nSQL and dbt projects get real `ref()` / `source()` lineage, shell scripts get\nfunction-level symbols, HTML pages contribute their `