CodeBurn is a local-first, zero-proxy terminal dashboard that tracks token usage, API costs, and performance metrics across 18 different AI coding tools — including Claude Code, Codex, Cursor, and many more. It runs entirely on your machine, reads session data directly from disk without requiring any API keys, and presents everything in a beautifully designed TUI dashboard. Whether you are a solo developer watching your budget or a team lead optimizing tooling costs, CodeBurn gives you the granular visibility you need.

Core Features

• 18 AI Coding Tools Supported — From Claude Code and GitHub Copilot to Zed AI, Cline, and KiloCode, CodeBurn auto-detects installed tools and reads their local session logs. No configuration needed.
• Zero API Keys, Zero Proxy — Everything runs locally. CodeBurn parses session metadata files from disk and prices each call using LiteLLM pricing model. No data ever leaves your machine.
• Interactive TUI Dashboard + macOS Menu Bar App — The terminal dashboard shows daily/monthly cost breakdowns by provider, model, task type, and project. A native macOS menubar app gives you real-time at-a-glance cost monitoring while you code.

What the Community Is Talking About

CodeBurn has an active GitHub community with 40 open issues. Here are some of the most discussed threads:

Issue #190 — "Is release v0.9.5 broken?" (16 comments)

A user reported that after upgrading to v0.9.5, the dashboard showed $0.0031 cost with only 2 calls and 0.0% cache hit despite having used AI coding tools all day. The discussion revealed a session parsing edge case that was quickly investigated. A maintainer responded acknowledging the issue and asking for log files to debug further. This kind of responsive issue triaging keeps a tooling project healthy.

Issue #159 — "Cursor Cost Tracking Not working" (11 comments)

Multiple users reported that Cursor's cost tracking displayed incorrect zero or near-zero data despite clear usage. The root cause was traced to changes in Cursor's session metadata format. The team implemented a fix and the discussion evolved into a broader conversation about handling session data versioning across tool updates. This is a great example of why open-source tooling projects need diverse real-world testing environments.

Issue #189 — "fix(codex): read complete session_meta line during discovery" (7 comments)

A contributor discovered that Codex CLI 0.128+ sessions with session_meta lines exceeding 16 KB were silently dropped from the dashboard. The root cause was a fixed 16 KB buffer read in src/providers/codex.ts. The PR discussion explored alternative approaches — whether to use streaming reads or dynamically grow the buffer — ultimately landing on a clean streaming line-reader implementation. This is a textbook example of a well-documented bug fix with thoughtful technical discussion.

Summary

CodeBurn fills a genuine gap in the AI coding toolchain: visibility. As developers increasingly rely on multiple AI assistants across different projects, understanding where your token budget goes is critical. With 6,100+ GitHub stars, active development, and a growing community of contributors, CodeBurn is quickly becoming an essential tool for any developer serious about AI coding costs. It is written in TypeScript, requires Node.js 20+, and installs via npm or Homebrew in seconds.

@getagentseal / codeburn · https://github.com/getagentseal/codeburn