Skip to content
JarenKemptonPublic

About

Local-first CLI and server that tracks AI usage and quota limits across your machines

Topics

Resources

Stars

0 stars

Watchers

0 watching

Forks

Latest commit

 

History

18 Commits

Folders and files

Repository files navigation

burn

Local-first CLI and server for AI usage and limits observability. Burn collects quota windows, token consumption, credits, and inference performance from the AI tools you already use — Claude Code, Codex, OpenRouter, LM Studio — normalizes them into versioned observations, and keeps durable history in SQLite on your own machines.

  • One executable, two roles: server (keeps history, serves the API) and collector (gathers usage on a machine and reports in). burn server run does both, so the server machine is covered automatically. An enrolled machine is called a node.
  • Collectors only make outbound connections; enrollment is a browser-approved device-authorization flow.
  • Tailscale Serve is the recommended transport (tailscale serve --bg 7337); LAN is an explicit opt-in.
  • Provider credentials never leave the collecting node; prompts and completions are never read into observations.

Install

curl -fsSL https://raw.githubusercontent.com/JarenKempton/burn/main/install.sh | bash

Installs the latest release binary to ~/.local/bin/burn (Linux/macOS, x64 and arm64). Update any time with burn update. From a source checkout: bun install && bun run build.

Quick start

# on the machine that keeps history — it collects on itself automatically
burn server run

# on every other machine that uses AI tools
burn collector run     # asks for the server URL on first run and walks you
                       # through browser-approved enrollment

burn providers add     # connect providers (local tools are auto-detected)
burn status            # machines, quota windows, usage

To reach the server from other machines, run tailscale serve --bg 7337 on it and enroll against https://<server>.<tailnet>.ts.net. burn server install writes systemd user units (Linux) or launchd agents (macOS) for install-and-forget operation.

For Claude Code 5-hour/weekly quota windows, run burn providers add claude_code — it offers to register Burn as your Claude Code statusline (the officially documented surface that carries rate-limit data).

Documentation

  • docs/spec.md — CLI/HTTP/SQLite contracts, auth and redaction invariants.
  • docs/fixtures/ — example observation payloads.
  • docs/research/ — provider research findings (issues #2, #3, #5, #6).

Development

bun test          # end-to-end: enrollment, ingestion idempotency, liveness
bunx tsc --noEmit # typecheck

About

Local-first CLI and server that tracks AI usage and quota limits across your machines

Topics

Resources

Stars

0 stars

Watchers

0 watching

Forks

Releases

Packages

Contributors

Languages