You're spending real money on AI coding tools every day. Exe Fuelbar shows you exactly where it goes — cost, tokens, models, projects, and whether the AI is getting it right the first time or burning through retry loops. One command, zero cloud, everything local.

npm install -g exe-fuelbar

What you get

Dashboard — Interactive TUI with gradient charts, responsive panels, keyboard navigation. Breaks down spend by day, project, model, activity type, tools, MCP servers, and shell commands. Auto-refreshes every 30 seconds.

One-shot rate — For every category that involves edits, Fuelbar detects edit/test/fix retry cycles and shows you the percentage of turns where the AI got it right on the first try. Coding at 90% means 9 out of 10 edits landed without retries.

Optimize — Scans your sessions and ~/.claude/ config for waste: re-read files, low read:edit ratios, uncapped bash output, unused MCP servers, ghost agents, bloated CLAUDE.md files. Hands back exact, copy-paste fixes. Grades your setup A through F.

Compare — Side-by-side model comparison on your own data. One-shot rate, retry rate, cost per edit, cache hit rate, delegation style, fast mode usage — broken down by task category.

Menubar — Native macOS app showing today's cost in your menu bar. Period switcher, trend/forecast/pulse insights, activity breakdowns, and export. One command to install.

Export — CSV and JSON export for any time period. Pipe --format json output to jq for scripting.

Supported tools

| Tool | Data source | Notes | |------|-------------|-------| | Claude Code | ~/.claude/projects/ | Full support | | Claude Desktop | ~/Library/Application Support/Claude/local-agent-mode-sessions/ | Full support | | Codex (OpenAI) | ~/.codex/sessions/ | Full support | | Cursor | SQLite (state.vscdb) | Auto mode estimated at Sonnet pricing | | cursor-agent | CLI sessions | Full support | | OpenCode | SQLite (~/.local/share/opencode/) | Subtask sessions excluded | | Pi | ~/.pi/agent/sessions/ | Full support | | OMP (Oh My Pi) | ~/.omp/agent/sessions/ | Full support | | GitHub Copilot | ~/.copilot/session-state/ | Output tokens only |

Auto-detected. If multiple tools have data, press p in the dashboard to toggle. Provider plugin system makes adding new tools straightforward — see src/providers/codex.ts for the pattern.

CLI reference

# Dashboard
exe-fuelbar                                    # interactive (default: 7 days)
exe-fuelbar today                              # today only
exe-fuelbar month                              # this month

# Reports
exe-fuelbar report -p 30days                   # rolling 30-day window
exe-fuelbar report -p all                      # everything on disk
exe-fuelbar report --from 2026-04-01 --to 2026-04-10
exe-fuelbar report --format json               # structured JSON to stdout
exe-fuelbar status                             # compact one-liner (today + month)

# Filter
exe-fuelbar report --provider claude           # single provider
exe-fuelbar report --project myapp             # project substring match
exe-fuelbar report --exclude tests             # exclude projects

# Tools
exe-fuelbar optimize                           # find waste, get fixes
exe-fuelbar optimize -p week                   # scope to last 7 days
exe-fuelbar compare                            # interactive model picker
exe-fuelbar export                             # CSV (today, 7d, 30d)
exe-fuelbar export -f json                     # JSON export

Dashboard keys: 1-5 switch periods (Today / 7d / 30d / Month / All). p toggle providers. c compare mode. o optimize view. q quit.

Flags work everywhere: --provider, --project, --exclude, --from, --to, and --format json combine freely across all commands.

Menubar app

exe-fuelbar menubar

Downloads, installs to ~/Applications, and launches. Re-run with --force to reinstall. Native Swift + SwiftUI — refreshes live via FSEvents plus a 15-second poll.

Compact mode drops decimals in the menubar (e.g. $110 instead of $110.20):

defaults write ExeFuelbarMenubar ExeFuelbarMenubarCompact -bool true

Configuration

Currency

exe-fuelbar currency GBP              # any ISO 4217 code (162 currencies)
exe-fuelbar currency --reset           # back to USD

Exchange rates from the European Central Bank via Frankfurter. Cached 24 hours. Applies everywhere: dashboard, menubar, exports.

Plans

Track spend against your subscription:

exe-fuelbar plan set claude-max        # $200/month
exe-fuelbar plan set claude-pro        # $20/month
exe-fuelbar plan set cursor-pro        # $20/month
exe-fuelbar plan set custom --monthly-usd 150 --provider claude
exe-fuelbar plan set none              # disable

Model aliases

If a model shows $0.00, your provider's model name doesn't match LiteLLM pricing data. Map it:

exe-fuelbar model-alias "my-proxy-model" "claude-opus-4-6"
exe-fuelbar model-alias --list
exe-fuelbar model-alias --remove "my-proxy-model"

Stored in ~/.config/exe-fuelbar/config.json. User aliases override built-ins.

Activity tracking

13 categories classified from tool usage patterns and keywords. No LLM calls, fully deterministic.

| Category | Trigger | |----------|---------| | Coding | Edit, Write tools | | Debugging | Error/fix keywords + tools | | Feature Dev | "add", "create", "implement" | | Refactoring | "refactor", "rename", "simplify" | | Testing | pytest, vitest, jest in Bash | | Exploration | Read, Grep, WebSearch without edits | | Planning | EnterPlanMode, TaskCreate | | Delegation | Agent tool spawns | | Git Ops | git push/commit/merge | | Build/Deploy | npm build, docker, pm2 | | Brainstorming | "brainstorm", "what if", "design" | | Conversation | No tools, pure text | | General | Uncategorized |

Reading the signals

| What you see | What it might mean | |---|---| | Cache hit < 80% | Unstable system prompt or caching not enabled | | Lots of Read calls per session | Agent re-reading files, missing context | | Low 1-shot rate (Coding 30%) | Agent struggling, retry loops | | Opus on small turns | Overpowered model for simple tasks | | Bash dominated by git status, ls | Agent exploring instead of executing | | Conversation category dominant | Agent talking instead of doing |

Starting points, not verdicts. A single experimental session with 60% cache hit is fine. That same number across weeks of work is a config issue.

How it works

Reads session data directly from disk. No wrapper, no proxy, no API keys needed. Pricing from LiteLLM (auto-cached 24h). Handles input, output, cache write, cache read, and web search costs. Deduplicates messages by API message ID (Claude), cumulative token cross-check (Codex), conversation/timestamp (Cursor), session+message ID (OpenCode), or responseId (Pi/OMP).

Environment variables:

| Variable | Description | |----------|-------------| | CLAUDE_CONFIG_DIR | Override Claude data directory (default: ~/.claude) | | CODEX_HOME | Override Codex data directory (default: ~/.codex) |

Contributing

Contributions welcome. The provider plugin system is the easiest entry point — each provider is a single file in src/providers/. See src/providers/codex.ts for the pattern.

src/
  cli.ts           Entry point (Commander.js)
  dashboard.tsx    TUI (Ink — React for terminals)
  parser.ts        Session reader, dedup, date filter
  models.ts        LiteLLM pricing engine
  classifier.ts   Activity classifier (13 categories)
  compare-stats.ts Model comparison engine
  export.ts        CSV/JSON export
  config.ts        Config management
  currency.ts      Currency conversion
  providers/       One file per supported tool

Star History

License

MIT

Built by Exe AI. Pricing data from LiteLLM. Exchange rates from Frankfurter. Inspired by ccusage and CodexBar.