Terse is an on-device token optimizer that reduces AI prompt and token costs by 40-70% for developers using AI coding tools. It runs entirely on your machine — your prompts, your code, and your files never leave your computer.
AI coding agents re-send a growing context window on every turn. The system prompt, the conversation history, and every file read or tool result get transmitted again and again — so token bills balloon long after you have stopped typing. Terse exists to give developers visibility and control over that cost, without changing the way they work.
Terse sits quietly alongside your AI coding tools and does three things:
Corrects typos, removes filler and hedging, and shortens verbose phrasing before your text is sent — cutting redundant tokens while preserving meaning.
Detects duplicate tool calls and redundant file reads in your agent sessions, so you can see exactly where tokens are being spent twice.
Measures per-turn token cost in real time, giving you a running picture of what each session is actually costing you.
The result is a workflow you already know — Claude Code, Cursor, ChatGPT, and the rest — with far fewer tokens flowing out the door. If you are new to the idea, our guide to what token optimization is explains the fundamentals.
Terse runs a compression pipeline over your prompt text at the point of entry — before the text enters the conversation history. Because agents carry context forward every turn, tokens saved once stay saved across every subsequent turn that includes that content. That compounding effect is what makes optimizing at entry so effective.
The pipeline is configurable through three optimization modes, so you choose the balance between compression and readability:
Alongside compression, Terse's agent monitor watches your sessions for duplicate tool calls and redundant file reads, and reports the token cost of each turn as it happens. Everything is computed locally — there is no cloud dependency for optimization, and no round-trip latency. Terse is built with Tauri, Rust, and Swift, and is available on both macOS and Windows.
For a deeper look at where the tokens actually go in an agent loop, see our breakdown of AI coding agent costs, or estimate your own spend with the token calculator.
Terse is privacy-first by design. Optimization runs 100% on-device: your prompt text, your source code, and your files are processed locally and never leave your machine. There is no cloud service reading your prompts, no telemetry pipeline shipping your code off to a server, and no external dependency required to compress a single token.
This matters because the text you send to AI coding tools is often your most sensitive material — proprietary source, internal docs, unreleased ideas. A token optimizer that routed that text through a remote service would trade one problem for another. Terse does not make that trade. If your machine is offline, Terse still works.
Terse is built for developers who work with AI coding agents every day and feel the token bill growing:
If you use Claude Code specifically, see our dedicated guide on Terse for Claude Code.
Terse works alongside the AI coding tools you already use. It detects when you are working in a supported tool and optimizes your prompts in place:
Terse offers a free tier so you can start cutting token costs at no cost, plus affordable paid plans for developers and teams who want the full optimization stack. See the pricing section for current plans.
We would love to hear from you — whether you have a question, a feature request, or just want to compare token-savings numbers. Reach us here:
Terse compresses prompts in real time, flags duplicate tool calls, and tracks every token — all on-device, with zero API calls. Cut 40-70% of your token costs across Claude Code, ChatGPT, and every AI tool you use.
Download Terse