codeburn logo

codeburn

Free

See where your AI coding tokens go. Interactive TUI dashboard for Claude Code, Codex, and Cursor cost observability.

Coding AssistantsFreeFree tier
#ai-coding-costs#token-usage-tracker#claude-code#cursor-ide#openai-codex#developer-tools#tui-dashboard#cost-observability#open-source#ai-agent-monitoring
Outputs: text
Type
Open Source
Company
getagentseal
codeburn screenshot

About codeburn

CodeBurn is a free, open-source, local-first tool that tracks AI coding token usage and cost across 36 tools and agents (Claude Code, Cursor, Codex, Gemini, Grok, GitHub Copilot, and more), broken down by model, project, and task. It provides an interactive terminal dashboard (TUI), a macOS menubar app, a web interface, and a GNOME Shell extension. Everything runs entirely on your machine—reading session files from disk with no API keys, accounts, or cloud services required. It includes a deterministic 13-category task classifier, one-shot success rate tracking, detailed cost breakdowns, an optimize command that identifies waste patterns, and export to CSV/JSON.

Key Features

13-Category Task Classifier: Deterministic classification of every coding session into Coding, Debugging, Feature Dev, Refactoring, Testing, Exploration, Planning, Delegation, Git Ops, Build/Deploy, Brainstorming, Conversation, or General — no LLM calls needed.
One-Shot Success Rate: Tracks how often each activity type completes on the first attempt vs. requiring retry loops, helping you spot where tokens are wasted on edit-test-fix cycles.
Multi-Provider Support: Reads session data from Claude Code, Claude Desktop, OpenAI Codex, Cursor, OpenCode, Pi, and GitHub Copilot with a plugin system for adding more.
Cost Breakdowns: Slice your spend by day, project, model (Opus/Sonnet/Haiku/GPT-5/GPT-4o/Gemini), activity type, core tools, shell commands, and MCP servers.
Optimize Command: Scans sessions and ~/.claude/ setup for waste patterns (re-read files, low read-to-edit ratio, unused MCP servers, ghost agents, bloated CLAUDE.md), returns copy-paste fixes and an A-F health grade.
Export & Integration: CSV and JSON multi-period export, --format json for pipeable structured output, --refresh for live monitoring, 162 currency support.
macOS Menubar App: Native Swift/SwiftUI app showing today's cost in your menu bar with popover for agent tabs, breakdowns, and optimize findings.

Pros & Cons

Pros
  • Free and open source (MIT license) – no paid tiers or feature gates.
  • Local-first ensures complete privacy; no data ever leaves your machine.
  • Supports a wide range of 36+ AI coding tools and agents.
  • Deterministic classification means no extra LLM calls for analysis, saving cost.
  • Detailed breakdowns by task, model, project, and tool provide actionable insights.
  • Optimize command offers concrete fixes to reduce token waste.
  • Multiple interfaces (TUI, menubar, web, GNOME) suit different workflows.
  • Active community with 8.9k stars and regular updates.
Cons
  • Requires Node.js 22.13+ and some local setup to get started.
  • Only works with supported tools that write session data to disk; unsupported tools won't show up.
  • No cloud-based multi-user team dashboard or synchronization built-in.
  • Windows support is not explicitly provided (no native app, but terminal TUI may work with Node.js).
  • Pricing accuracy depends on LiteLLM's daily refresh and may lag behind actual API price changes.

Best For

Developers using Claude Code: Track exactly how many tokens each coding session consumes, broken down by task type, so you can understand where your Max plan budget is going.Teams using Cursor or Codex: Get per-project cost breakdowns across multiple AI coding tools, with the ability to filter by provider and export to CSV for team reporting.Engineering managers: Justify AI tool spend with concrete data on which activities, models, and projects consume the most resources, and identify waste patterns across the team.Cost-conscious indie developers: Use the optimize command to find and fix waste patterns like re-read files, unused MCP servers, and bloated context files that inflate your token usage.

Alternatives to codeburn