AgentFit

Fitness tracker dashboard for AI coding agents. Reads your local Claude Code and Codex conversation logs, syncs them into a SQLite database, and presents rich usage analytics — cost, tokens, tool usage, productivity patterns, and more.

Install

Option 1: Desktop App (recommended)

Download a pre-built installer from the Releases page:

• macOS: AgentFit-x.x.x.dmg (Intel) or AgentFit-x.x.x-arm64.dmg (Apple Silicon)
• Windows: AgentFit-x.x.x.exe

macOS note: If you see "AgentFit Not Opened" on first launch, go to System Settings > Privacy & Security, scroll down, and click Open Anyway. Or run xattr -cr /Applications/AgentFit.app in Terminal.

Option 2: One-liner

curl -fsSL https://raw.githubusercontent.com/harrywang/agentfit/main/setup.sh | bash

Option 3: npx

npx agentfit

Option 4: Manual

git clone https://github.com/harrywang/agentfit.git
cd agentfit
npm install
echo 'DATABASE_URL="file:./agentfit.db"' > .env
npx prisma migrate deploy
npm run build
npm start

Open http://localhost:3000. The app auto-syncs your Claude Code (~/.claude/projects/) and Codex (~/.codex/sessions/) logs on first load.

Requirements: Node.js 20+ (Options 2–4)

The CRAFT Framework

AgentFit scores your AI coding proficiency using CRAFT — a Human-AI coding proficiency framework by Harry Wang. All metrics are computed from your local conversation logs.

| | Dimension | What it measures | Key metrics | |---|---|---|---| | C | Context | How effectively you engineer context for the AI | CLAUDE.md usage, memory writes, cache hit rate | | R | Reach | How broadly you leverage available capabilities | Tool diversity, subagent usage, skill adoption | | A | Autonomy | How independently the agent works for you | Message ratio, interruption rate, delegation | | F | Flow | How consistently you maintain a coding rhythm | Streak length, daily consistency, active days | | T | Throughput | How much output you get for your investment | Cost efficiency, output volume, error rate |

Inspired by DORA Metrics and Microsoft's SPACE framework. Behavioral signals from your logs — no surveys, no guesswork. Each dimension is scored 0–100.

Features

• Dashboard — overview stats (cost, tokens, sessions, projects, messages, tool calls, duration)
• CRAFT Coach — fitness score, achievements, and actionable improvement tips
• Daily Usage — daily cost and activity charts
• Token Breakdown — pie chart + stacked area chart of token types
• Tool Usage — top tools by invocation count
• Projects — per-project breakdown with top tools and sessions
• Sessions — individual session details with chat logs and tool flow graphs
• Personality Fit — MBTI-style behavioral analysis
• Command Usage — slash command pattern tracking
• Images — screenshot analysis across sessions
• Community Plugins — extensible analysis views

Development

Build the desktop app locally:

npm run electron:build:mac   # Mac (.dmg)
npm run electron:build:win   # Windows (.exe)

npm run dev          # Start dev server (Turbopack)
npm run build        # Production build
npm run lint         # ESLint
npm run format       # Prettier
npm run typecheck    # TypeScript check
npm test             # Run tests

Community Plugins

See CONTRIBUTING.md for the full guide. Quick version:

1. Create plugins/<your-slug>/manifest.ts and component.tsx
2. Register in plugins/index.ts
3. Add tests in component.test.tsx
4. Submit a PR

Configuration

| Variable | Default | Description | |----------|---------|-------------| | PORT or AGENTFIT_PORT | 3000 | Server port | | DATABASE_URL | file:./agentfit.db | SQLite database path |

Credits

• Logo: Robot SVG from SVG Repo (CC0 License)

License

MIT