Codeburn — AI Coding Cost Observability Dashboard

Introduction

Codeburn reads session logs from your AI coding tools and presents token usage, cost breakdowns, and trend data in a rich terminal UI. It answers the question every developer using AI assistants eventually asks: where are all these tokens going?

What Codeburn Does

• Parses session data from Claude Code, Codex CLI, and Cursor to extract token counts and costs
• Displays real-time and historical cost breakdowns in an interactive TUI dashboard
• Groups spending by project, session, model, and time period
• Highlights expensive operations and sessions that consumed the most tokens
• Exports reports as JSON or CSV for further analysis

Architecture Overview

Codeburn is a TypeScript application that runs as a local TUI using the Ink framework. It scans known log directories for each supported tool, parses session metadata, and calculates costs using current model pricing tables. The dashboard renders charts, tables, and summaries in the terminal using box-drawing characters and ANSI colors. No data leaves your machine.

Self-Hosting & Configuration

• Install globally with npm install -g codeburn or run via npx codeburn
• Automatically discovers Claude Code, Codex, and Cursor log directories
• Custom log paths can be configured in ~/.codeburn/config.json
• Pricing tables update from the bundled defaults; override with custom rates if needed
• Runs entirely offline — no network calls, no telemetry

Key Features

• Unified view across multiple AI coding tools in one dashboard
• Interactive TUI with drill-down from summary to individual sessions
• Cost trend visualization to track spending over time
• Zero network dependency; all analysis runs locally
• Open source under MIT license

Comparison with Similar Tools

• Claude Code /cost — shows current session only; Codeburn aggregates across all sessions and tools
• Quotio — macOS menu bar quota tracker; Codeburn provides deeper per-session analysis in the terminal
• Helicone / Langfuse — cloud-based LLM observability; Codeburn is local-only and focuses on developer cost
• TokenCost — library for programmatic cost calculation; Codeburn is an end-user dashboard

FAQ

Q: Does Codeburn send data anywhere? A: No. All processing happens locally on your machine. There are no network calls.

Q: Which AI tools are supported? A: Claude Code, Codex CLI, and Cursor. More backends are planned.

Q: Can I track team-wide costs? A: Codeburn is designed for individual developers. Export data as JSON and aggregate externally for team use.

Q: Does it work on Windows? A: Yes, anywhere Node.js runs — macOS, Linux, and Windows.

Sources

• https://github.com/getagentseal/codeburn