repowiseEvery question your agent asks about your repo has an answer that could have been computed ahead of time. Who calls this function? What breaks if I change it? Why is it written this way? Which of these files is actually dangerous? Instead, agents rediscover it from scratch on every task: grep, read, re-read, forget.
Repowise computes those answers once and keeps them current on every commit.
Your agent reads the answer instead of the codebase, through
ten task-shaped MCP tools, and the same index gives your
team a defect-validated health score, change-risk scoring on every PR, and a
local dashboard for all of it. One pip install, no cloud, and the first index
needs no API key.
Start here
Quickstart
Install repowise, index your first repo, and connect Claude Code in about a minute.
Set up with your agent
Skip the manual setup: a copy-pasteable prompt your agent follows start to finish, no API key anywhere in the flow.
The ten MCP tools
What each tool answers and when your agent should reach for it.
What is repowise
The five layers, who it is for, and how it differs from plain LLM coding.
What one index builds
Five layers, built in a single pass and kept in sync on every commit. Each is queryable from the CLI, the MCP tools, and the local dashboard.
| Layer | What it gives you |
|---|---|
| Graph | Dependency graph across 19 languages, file and symbol nodes, call resolution stamped with one of 29 confidence-scored origins, Leiden communities, PageRank, and execution-flow tracing. |
| Git | Hotspots from decayed churn, ownership percentages, co-change pairs that reveal coupling no import graph shows, bus factor, and which files actually get bug-fixed. |
| Docs | A wiki page per module and notable file, rebuilt incrementally on every commit, with freshness and confidence scoring. Rendered from your code's structure with no API key, or upgraded to model-written prose page by page. |
| Decisions | Architectural decisions mined from git, captured by hand, or extracted from inline markers, each traced to a verbatim source span and linked to the files it governs. |
| Code health | A 1 to 10 score per file from 49 deterministic detectors across defect risk, maintainability, and performance, with a concrete refactoring plan attached. Zero LLM calls, under 30 seconds. |
The whole wiki is generated with no LLM, then upgraded on demand.
repowise init --no-prose builds every layer above and renders each wiki page
from your code's structure, with no API key and no spend. Convert any part of
it to model-written prose whenever you want, one page or a ranked slice at a
time, with the cost shown before you confirm.
What the measurements say
Three numbers worth knowing before you spend an afternoon on this, each with its sample and its test. The one thing we will not do is quote a figure without the caveat attached.
- 0.876 file coverage at finding the code a question is about, against the next tool's 0.610. Measured on a 42-instance split sealed before any of the work started, graded deterministically with no LLM judge (p = 0.00004).
- 31.6% fewer output tokens written by the agent itself, across 43
questions on
django/djangounder Codex, reached in 3.8 tool calls where a bare agent needed 7.2 (p < 0.0001). CodeGraph is a genuine second here at 24.4%: we lead a field in which more than one tool works. - 393 tokens to load a commit's context through
get_context, against 13,984 for reading every changed file, counted with deterministictiktokenover 30 commits. That is 35.6x pooled, and it is the easy number: it measures one payload, not a session.
Defect risk is validated separately at ROC AUC 0.737 across 21 repositories and 9 languages, scored at a historical commit with nothing after it feeding the score.
We also publish the row we lose. Repowise is the slowest indexer in the
field, 22x CodeGraph on django/django, because the same pass that builds a
call graph also builds four more layers. It is a one-time cost, and updates
after the first index are incremental.
Method, raw cells, pre-registrations, and the runs we invalidated: repowise.dev/benchmarks.
Hosted or self-hosted
Same engine, same ten tools, same MCP protocol. The choice is who runs the indexer.
# Self-hosted: runs entirely on your machine, your code never leaves it
pip install repowise
cd your-project
repowise initHosted adds managed indexing, push-sync through a GitHub App, private repos, AI docs and chat, and the team surfaces. Sign in at repowise.dev and index a repo; there is nothing to install.
Hosted vs self-hosted
What each one runs, what it costs, and which fits your team.
Security suite
CVE scanning with reachability triage, secrets across full git history, SBOM and VEX, nightly refresh, audit trail.
Agent plugins
The Claude Code and Codex plugins: MCP server, hooks, six skills, and ten slash commands in one install.
VS Code extension
Health signals, refactoring CodeLens, and branch risk in the editor, plus one-install MCP for your agent.
CLI reference
Every repowise command grouped by purpose, verified against source.
Where to find help
- Discord: discord.gg/cQVpuDB6rh
- GitHub: github.com/repowise-dev/repowise
- Email: hello@repowise.dev
- Stuck on install? Run
repowise doctor, or read Troubleshooting.