Directory · Subagents

agentburn

Local, zero-dependency profiler that shows where Claude Code and other agents burn tokens, usage windows, and money.

★ 113 stars · last updated 2026-10-12 · Free, open source · View on GitHub

Best for

• Claude Code Pro/Max users finding which 5-hour window hit the usage limit
• Per-token users tracking overnight cron, gateway, and subagent spend
• Getting dry-run config fixes like removing unused MCP servers or trimming CLAUDE.md
• Live usage window status in the Claude Code statusLine

Install

uvx agentburn

Our take

Python 3.9+ CLI with no dependencies that reads local agent logs read-only and sends nothing off the machine; supports Claude Code, Codex CLI, Gemini CLI, opencode, OpenClaw, and Hermes Agent. It reports tokens rather than dollars for Claude Code, derives usage ceilings from recorded cut-offs instead of guessing, and deliberately has no auto-apply for fixes. It fits the Subagents category loosely, since subagent cost rollups are one feature of a broader usage profiler.

Similar subagents

• Awesome Claude Code Subagents (VoltAgent): A curated collection of 161+ Claude Code subagent definitions across 10 categories, installable as plugins or individual agent files.
• claude-devtools: Desktop and web app that reads local Claude Code session logs to show tool calls, token usage, subagent trees, and context.
• AI Marketing Suite for Claude Code: Claude Code skills and parallel subagents for marketing audits, copywriting, email sequences, ad copy, and client-ready reports.
• pentest-ai-agents: 50 Claude Code subagents turning Claude into an offensive security research assistant for authorized penetration testing.
• Translate Book: Agent skill for Claude Code, Codex, and OpenClaw that translates whole PDF/DOCX/EPUB books using parallel subagents per chunk.
• Claude Code Sub-Agents (lst97): A collection of 33 specialized Claude Code subagents for full-stack development, plus an agent-organizer for multi-agent orchestration.