tokensitter
The Tokensitter toolkit

Your tokens.
Better looked after.

Less unnecessary work. Clearer choices. A readable account of what changed. Explore the technique, quality checks and evidence behind each feature.

Add to Claude CodeOpenAI Codex: coming soon

Smart routing

Choose a model at each step of API-key work. Keep the requested model when the request needs it, and account for cache costs before switching.

Technique · Quality · Comparison · Evidence

Efficient file reading

Ask a cheaper model to read a file or log, then return a short answer with path-and-line excerpts. Use exact symbol reads when a bounded source slice is enough.

Technique · Quality · Comparison · Evidence

Shorter replies

A short instruction on each prompt asks for more focused output. Ultra is the Claude Code plugin default; Off and Lite give you control.

Technique · Quality · Comparison · Evidence

Lean code

A short local instruction asks the coding agent to plan first and change only what the task needs. Lite is the plugin default.

Technique · Quality · Comparison · Evidence

Skills & MCP context

See the context added by skills, plugins and MCP tool definitions. Then choose what deserves a place in every conversation.

Technique · Quality · Comparison · Evidence

Savings & measurement

Read what is observed, what is calculated, and what remains an estimate. Cash cost, plan usage and potential savings answer different questions.

Technique · Quality · Comparison · Evidence

Token compression

Deterministic rules shrink new tool results at the API relay. No extra model call is needed for this transformation.

Technique · Quality · Comparison · Evidence

Read less

Local hooks turn noisy command output and large structured files into a compact view while keeping a route to the original.

Technique · Quality · Comparison · Evidence

Cheaper helpers

Use a less expensive helper for clearly read-only work while keeping edits and ambiguous tasks out of that path.

Technique · Quality · Comparison · Evidence

Cache keep-alive

An experimental local hook can read a large cached prefix before it expires, so a later return may avoid rebuilding it.

Technique · Quality · Comparison · Evidence

Reversible context

Replace eligible old tool results with short stubs, keep the originals encrypted per key, and retrieve them when the model needs them.

Technique · Quality · Comparison · Evidence
Proof before promises. Lever results are measured separately and are not additive: do not sum them into one combined saving. Each page separates API relay and local plugin scope, names the experiment, and keeps quality outcomes and limits beside the figures. Evidence comes from our lab reports (snapshot 27 Oct 2026); no competitor head-to-head is claimed.

Read the small print.
It should be worth reading.

Baselines, quality, fees and uncertainty belong beside the results.

Our measurement standard