What is it?
What it is
An always-on token‑efficiency skill for coding agents (Claude Code, Codex, Cursor, Windsurf, Cline) that trims token usage across replies, docs, tests, code, context, and tool use while preserving correctness.
Why it exists
To reduce LLM API costs and output verbosity for coding agents without sacrificing correctness, delivering ~31% lower bills on average.
Who should use it
Developers and teams using Claude Code, Codex, Cursor, Windsurf, or Cline who want to reduce token consumption, lower costs, and keep output concise without sacrificing correctness.
Who should avoid it
Users who rely on verbose, explanatory output for training or debugging, or who do not use the supported AI agents and cannot integrate the always‑on shell skill.
How it works
A quick walkthrough in plain English
How token-diet works
Step 1 of 3
You interact with it
Open token-diet, send a request, or connect it to your stack.
Features
Advantages
- Reduces output tokens by 30%–81% depending on session type
- Lowers bill by ~31% on average (−17% to −54%)
- Maintains correctness and critical test coverage in all runs
- No per‑message command needed; works automatically every session
- Works across multiple popular AI coding agents
- Simple one‑liner install script
- Configurable verbosity levels let users balance brevity and clarity
- Preserves essential information while removing filler and narration
Disadvantages
- Requires installation and shell access to run the installer
- Effectiveness is lower for read‑heavy sessions where file‑reading dominates
- Savings vary widely by session type (output‑heavy vs code change vs comprehension)
- Relies on agent‑specific integration points (SessionStart hook, AGENTS.md, rule files)
- Ultra level may remove explanatory detail that some users find helpful
- No explicit license specified (license: null) may concern some users
Installation
FAQ
How do I install token-diet?
Run the one‑liner installer: `curl -fsSL https://raw.githubusercontent.com/Kulaxyz/token-diet/main/install.sh | bash`. You can also pass options like `--ultra` for telegraphic mode or `--uninstall` to remove it. If you prefer cloning, run `./install.sh` after cloning the repo.
What are the different levels of token-diet and what do they change?
`on` (default) applies all optimisations. `lite` keeps only concise communication and artifacts. `ultra` uses telegraphic chat while still keeping code, tests, and docs precise. `off` disables the skill entirely.
How does token-diet trim token usage?
It removes unnecessary preambles and postambles from replies, shortens docs and comments, limits tests to essential paths, reads only needed file lines, batches tool calls, and delegates broad searches to cheaper models. All of this keeps output concise without sacrificing correctness.
Can I target a specific agent or install it only for a project?
Yes. Use the `-a` flag to target an agent (`claude`, `codex`, `cursor`, `windsurf`, `cline`, or `all`). Add `--project` to install into the current repository instead of globally.
How do I enable or disable token-diet during a session?
The skill is always‑on by default, but you can toggle it on demand with the command `/token-diet [on|lite|ultra|off]` within a session.
Featured in Videos
YouTube tutorials and walkthroughs for token-diet
t5
70,223,459 views
Starts at 00:00
Alternatives
Similar projects ranked by category, topics, and text overlap.
