What is Repowise?
Repowise indexes a repository once on your machine: a call graph across 28 parsed languages, git history, ownership, tests, dead code, a 1 to 10 health score for every file and decisions mined from ADRs, comments, commits and agent sessions. Coding agents read it through ten default MCP tools (seven more are opt-in), such as get_answer, get_context, get_risk and get_change_risk; people read the same index in a local dashboard, a VS Code extension and a CLI.
The graph, risk, health, test and dead-code layers make no LLM calls and need no API key. Generated prose, decision extraction and refactoring code are optional and use your own model provider or Ollama. Other pieces include repowise distill, which trims test, install and git output before the agent reads it, generated CLAUDE.md and AGENTS.md files, change-risk scores for CI and a free PR bot.
The code is AGPL-3.0, written in Python and installed from PyPI. The same team runs a hosted version at repowise.dev with paid plans, and sells commercial licences for self-hosting without AGPL terms. Version 0.60.1 shipped on 2 November 2026.
Is Repowise worth it?
Repowise is a code index that does two jobs from one build: it answers an agent's questions about a repository, and it scores the code for the humans who maintain it. Most context tools for agents stop at search or a call graph; this one adds git history, ownership, test reach and change risk.
The maintainers publish the numbers behind the claim, and they are their own. On 43 questions about django/django, an agent with Repowise made 3.8 tool calls instead of 7.2, opened 3.0 files instead of 7.2 and wrote 31.6% less output. On 42 sealed ContextBench cases, get_answer found 0.876 of the files a fix needed against 0.610 for CodeGraph. The health score found 2.3 times as many later-buggy files as CodeScene at the same review budget across 2,770 files. All of these are vendor benchmarks, with methods in docs/BENCHMARKS.md; we found no independent replication.
The same benchmark page is honest about the cost. Building the call graph is fast and light (2.77 seconds and 75 MB of memory, the lowest of five tools on all 35 repositories tested), but building the full index of django took 366.8 seconds against 16.4 for CodeGraph, because Repowise also builds the history, wiki, decisions and health layers. If a call graph is all you need, CodeGraph or Serena is the lighter choice.
The things to check before adopting it: the CLI sends anonymous telemetry by default (command names, flag names and coarse file-count ranges, never code or paths), switched off with repowise telemetry disable or DO_NOT_TRACK=1. The AGPL-3.0 licence costs nothing for internal use, but embedding Repowise in a product you ship needs a commercial licence. And the project moves fast: version 0.60.1 seven months after the repository was created, with 467 open issues and pull requests.
The health score is where it differs most from agent-context rivals. CodeScene flags about 27 files where Repowise flags 132 on the same benchmark, with higher precision (0.636 against 0.580), so a team that wants a short list to act on may find the incumbent's threshold more useful.
Where it earns its place
A backend engineer joining a 500,000-line Python monolith
Get Claude Code to answer questions about the code without reading dozens of files each time
repowise init --yes --no-prose, then repowise serve; Claude Code calls get_overview and get_answer instead of grepping. The vendor's django run cut tool calls from 7.2 to 3.8 per question. No API key or spend for the index itself.
An engineering manager with a 12-person team and a backlog of legacy code
Decide which files to refactor first this quarter
repowise next ranks files by impact and effort from churn, fan-in, coverage and bug history, and repowise health --refactoring-targets gives concrete plans. Self-hosted it is free; hosted Teams is $20 a seat a month, or $240 a month for 12 people.
A platform team running payment services across three repositories
Catch API changes that would break another team's service before merge
Add the repositories to a workspace and run repowise workspace check in CI; it matches HTTP, gRPC and event contracts across repositories and names the consumer files a breaking change affects. No LLM calls, so the same diff always gets the same answer.
Before you commit
- Setup
- uv tool install repowise (or pipx or pip), then repowise init --yes --no-prose in the repository and repowise serve for the dashboard and MCP server; the dashboard needs Node 20 or later. init wires Claude Code automatically; other agents get a config snippet.
- Integrations
- Ten default MCP tools plus seven opt-in, full support for Claude Code and Codex CLI, good support for VS Code, Cursor, OpenCode, Hermes, GitHub Copilot CLI and Kiro; a VS Code extension, a GitHub Action, a GitLab CI template, SARIF output and a free GitHub PR bot.
- Data and privacy
- Self-hosted, the index stays on your machine and raw source is not stored. Anonymous CLI telemetry is on by default and kept 90 days; optional prose features send code-derived prompts to your own LLM provider, or nothing leaves the machine with Ollama. The hosted service indexes your code on its platform.
- Cost at scale
- Self-hosting is free; the vendor measured 75 MB of memory to build the call graph, and indexed the 15-million-line dotnet/runtime on one laptop. Hosted Pro is $15 a month for 5 private repos ($20 for new subscribers from 1 December), Teams $20 a seat a month with a 3-seat minimum and 25 pooled repos; commercial self-hosted licences are priced per seat, repository or enterprise and not published.
- Limits
- Full indexing with every layer is much slower than a graph-only tool; 6 lightweight and 8 structural languages get import graphs or git history only; benchmarks are run by the vendor; 467 open issues and pull requests.
What’s good
- One index serves the agent over MCP and people in a dashboard
- Graph, risk, health and dead code need no API key or LLM calls
- Published, reproducible benchmarks, including the rows it loses
- Reversible command-output trimming with repowise distill
- Cross-repo contract map and breaking-change guard
What’s not
- Full index is slow: 366.8 s on django against 16.4 s for CodeGraph
- Anonymous CLI telemetry is on by default
- AGPL-3.0; shipping it inside a product needs a commercial licence
- 467 open issues and pull requests; interfaces change often
Best for
Teams using Claude Code, Codex or Cursor on a large or unfamiliar repository who also want code health, change risk and test impact from the same local index.
Repowise alternatives
Repowise: common questions
- Is Repowise worth it?
- Repowise gives a coding agent a map of the repository and gives the team a code health tool from the same index, at the cost of a slow first build and a large, fast-moving surface. We rate it 4.1 out of 5.
- How much does Repowise cost?
- Free to self-host under AGPL-3.0. Hosted repowise.dev: Free for 2 public repos; Pro $15/month for 5 private repos ($20 for new subscribers from 1 December); Teams $20/seat/month, minimum 3 seats; Enterprise custom.
- What are the downsides of Repowise?
- Full index is slow: 366.8 s on django against 16.4 s for CodeGraph; anonymous CLI telemetry is on by default; aGPL-3.0; shipping it inside a product needs a commercial licence; 467 open issues and pull requests; interfaces change often.
- Who is Repowise best for?
- Teams using Claude Code, Codex or Cursor on a large or unfamiliar repository who also want code health, change risk and test impact from the same local index.
- How much does Repowise cost for a team or in production?
- Self-hosting is free; the vendor measured 75 MB of memory to build the call graph, and indexed the 15-million-line dotnet/runtime on one laptop. Hosted Pro is $15 a month for 5 private repos ($20 for new subscribers from 1 December), Teams $20 a seat a month with a 3-seat minimum and 25 pooled repos; commercial self-hosted licences are priced per seat, repository or enterprise and not published.
- Is Repowise safe for company data?
- Self-hosted, the index stays on your machine and raw source is not stored. Anonymous CLI telemetry is on by default and kept 90 days; optional prose features send code-derived prompts to your own LLM provider, or nothing leaves the machine with Ollama. The hosted service indexes your code on its platform.
- What are the best Repowise alternatives?
- The closest alternatives we have reviewed are Graphify.
Answers checked Nov 5, 2026.
Make Repowise? Show this rating on your site
