llm-spend

Local LLM cost dashboard for Claude Code. One command, zero setup.

npx llm-spend

Opens a dashboard at http://localhost:3000 showing exactly where your Claude tokens and dollars are going — by model, session, project, and individual request.

Features

• Zero setup — auto-reads ~/.claude/projects/ on startup, no config required
• Overview — daily spend chart, cost-by-model donut, 5 stat cards with tooltips
• Actionable insights — 8 expandable insight cards (cache efficiency, model mix, session efficiency, spend trajectory, monthly projection, and more)
• Projects breakdown — cost ranked by project path
• Sessions tab — top sessions by cost with cost-per-request efficiency metric
• Top Requests by Cost — 15 most expensive individual API calls, with the original prompt shown inline (collapsible)
• Date range filters — Today / 7D / 30D / All, with live chart refresh
• Privacy-first — server binds to 127.0.0.1 only, no telemetry, no external calls

Quickstart

# One-off (no install)
npx llm-spend

# Or install globally
npm install -g llm-spend
llm-spend

CLI Options

llm-spend [options]

Options:
  -p, --port <number>      Port to listen on (default: 3000)
  --no-open                Do not open browser automatically
  --claude-dir <path>      Path to Claude data directory (default: ~/.claude)
  -V, --version            Output version number
  -h, --help               Display this help

Supported Models

Anthropic: claude-opus-4-6, claude-sonnet-4-6, claude-haiku-4-6 (and 4.5, 4, 3.7, 3.5, 3 variants) OpenAI: gpt-4o, gpt-4o-mini, gpt-4-turbo, gpt-4, gpt-3.5-turbo, o1, o1-mini, o3-mini, o4-mini

Unknown models are tracked but show $0 cost.

Privacy

• Server binds to 127.0.0.1 — not accessible from other machines
• No prompts or responses are stored — only token counts, model name, timestamps, and computed cost
• No telemetry or external HTTP calls whatsoever
• All data is read directly from your local ~/.claude/ directory

License

MIT