First public release.
The wheel arrives complete — everything below ships in this release, free and open source under MIT. Available on macOS, Windows, and Linux.
A health read at a glance
- Project health score — one number, built from independent signals (bus factor, commit quality, TODO density, duplication, churn, and dependency health), each shown on its own so you can see what's pulling the score up or down.
- History and trends — every analysis is snapshotted automatically, so you can watch health, TODOs, bus factor, coverage, and more move over time.
- Then-vs-now comparison — pick two points in your history and see exactly what changed.
- Onboard and Monitor modes — a "where do I start?" view for getting into an unfamiliar codebase, and a steady-state view for keeping an eye on one you know.
- Shareable reports — export the whole picture to a self-contained HTML or PDF file.
The people and the history
- Contributors and activity — who commits, when, and how much; a commit timeline and a bus-factor read on how concentrated your knowledge is.
- Teams and velocity — group contributors into teams and track each team's output over time.
- Commit quality — how many commits follow a clear convention, with examples.
- Stale-branch and quiet-contributor alerts — surfaces branches gone cold and risky files owned mainly by someone who's stopped contributing.
Code quality, in depth
- Complexity — cyclomatic complexity per function across 11 languages, with the hotspots ranked and a per-file trend.
- Tech debt — an estimate per file, cross-checked against how often each file actually needed hotfixes.
- Duplicates, TODOs, and large files — the noisy spots, filterable by severity and package.
- Dead code — unused imports, exports, and assets for JavaScript, TypeScript, Python, PHP, and Java.
- Test coverage — reads LCOV, Jest, Cobertura, JaCoCo, and Go cover profiles, no setup required.
- File treemap — the whole codebase as a heatmap: size by lines, color by churn.
Dependencies and risk
- Security and outdated packages — vulnerability and outdated checks across seven ecosystems (npm, pip, Cargo, Go, Ruby, PHP, Java), with circular dependencies surfaced in Architecture.
The wheel — set intent, review, verify, dispatch
- Goals — set what you're steering toward in plain language ("fewer useEffects", "no file over 500 lines", "keep coverage above 80%"); Reikon turns it into a deterministic rule once and tracks current vs. baseline on every analysis, with an improving / holding / regressing status.
- Review — staged/unstaged or branch (base...HEAD) diff, unified or split, with an inline annotation layer (fix / explain / review), threaded replies, a per-diff health summary, a suggested reviewer, and a "safe to refactor?" read on any file. Notes follow the code when it's edited, and resolved notes are kept as durable review history.
- Verify — your coverage report turned into a ranked "test these first" list: the riskiest files (complex, churning, single-owner) with little or no tests.
- Architecture — name your layers (or let Reikon suggest them from your code) and which may import which; every boundary-crossing import becomes a finding, and each new crossing is a recorded decision — allow it, or keep it as a violation to fix.
- Dispatch — from any note or goal, hit Copy prompt and Reikon composes the kickoff for your agent — or, with MCP connected, just tell your agent to work Reikon's queue and it pulls the work itself. The fix is made in the agent's own environment and the diff returns to Review for you to accept or send back. Reikon emits the instruction and reviews the result — it never applies a fix, and a returned fix only counts once there's a real change in the working tree.
- Several repos at once — open up to three in tabs; a dispatched fix in a background repo pings its tab when it lands, with a dispatch ledger across every open repo.
AI — optional, keyless
- Point Reikon at a coding-agent CLI you already have — Claude Code, Codex, or Gemini — and it will explain your health score, categorize your TODOs, suggest cleaner commit messages, write a standup summary, and answer questions about the project. Reikon never holds a key or makes the call itself; the CLI owns its own auth. Fully useful without it.
Works with your AI agent
- MCP server, in the box — coding agents — Claude Code, Cursor, Windsurf, Gemini, GitHub Copilot, or anything MCP-compatible — can pull deep project context (health, file risk, diff review, complexity hotspots, dead code) in a single call. Agents can also flag risky hunks (server-verified against Reikon's own thresholds, so they can't cry wolf), reply to your notes, and return dispatched fixes for your review — the write tools touch review notes only, never your source.
Command line and command palette
- rei CLI — run the same analysis headlessly (JSON / HTML / summary), watch for new commits, and gate CI on the result. The review loop works from the terminal too: rei review lists and works your flags (reply / resolve / request-fix / send-back), rei status is the dispatch ledger across your repos, and rei mcp hosts the MCP server with no desktop app.
- Command palette — jump to any file, contributor, TODO, branch, or action from one search box.
Earlier pre-release iterations live in the git history — releases are tagged on GitHub.