What it does
Token Optimizer is a tool for reducing the tokens an AI coding assistant wastes. After you install it and run an audit once, hooks work in the background: they checkpoint work before compaction and restore it afterwards, compress repeated file reads, command output and search results, and archive large tool results so they can be expanded on demand. A quality score with letter grades tracks how healthy a session is. The audit looks at bloated configuration files, unused skills, stale memory and model choice, and commands such as quick, route, doctor, savings and report cover health checks, model suggestions and cost breakdowns. A local dashboard regenerates after each session with per-turn token breakdowns, cost across pricing tiers, cache analysis and a savings tracker. The site says nothing leaves your machine and that savings are labelled as measured or estimated. It installs as a plugin for Claude Code, and the page gives install paths for Codex, OpenCode, OpenClaw, Hermes, Pi Coding Agent, GitHub Copilot, Cursor and Google Antigravity. Windows users are told to use the plugin install only. The repository lists its license as Other.
Capabilities
- Checkpoints before compaction and restores afterwards
- Compresses repeated file reads, command output and search results
- Session quality scoring with letter grades
- Local dashboard of tokens, cost, cache and savings
- Audit of configs, skills, memory and model routing
- Install paths for Claude Code, Codex, Cursor and others
When to use it
- Your coding agent's context fills up and compaction loses work
- You want to see where token spend goes across sessions
- You want to find unused skills or bloated config files in your agent setup
Alternatives to explore
- Headroom
- RTK
- context-mode
