Claude Code

Cut Your Claude Code Bill by 40–89%

Terse auto-detects Claude Code sessions, compresses every prompt, catches duplicate tool calls, and generates CLAUDE.md rules — so you spend less and ship faster.

Start 30-Day Free Trial See How It Works
89%CLI noise reduction
40–70%prompt compression
Autosession detection
0data leaves device

How Terse Works with Claude Code

A 2-hour Claude Code session generates ~210,000 tokens of output noise. Terse intercepts, compresses, and monitors every turn — automatically.

Terminal — Claude Code Session
Terse Active
$ claude
Claude Code v1.2.4 — type /help for commands
● Terse detected — monitoring session (Opus · $15/M tokens)
> refactor the authetication middlewre to use asynchrnous handlers
Terse optimized: authetication middlewre asynchrnous → authentication middleware asynchronous · saved 3 tokens · prevented 2 retry tool calls
Reading: src/middleware/authentication.ts [847 tokens]
⚠ Terse: duplicate file read detected (3rd time this session) · flagging for CLAUDE.md rule
23.4Ktokens used
210Kwithout Terse
89%compressed
$0.31session cost
⚡

Auto Session Detection

Terse scans processes every 5 seconds. The moment Claude Code launches, a floating monitor bar appears — no setup, no config required.

Automatic
🗜️

Prompt Compression

Every prompt you type goes through a 7-stage pipeline: typo fix, whitespace normalization, filler removal, pattern optimization, and more. Code blocks are always protected.

7-Stage Pipeline
🔍

Duplicate Tool Call Detection

Terse watches for repeated Read, Bash, and Search calls within a session. Each duplicate is flagged in real time before it burns tokens.

Agent Monitor
📁

Redundant File Read Flagging

When Claude Code reads the same file multiple times — a very common pattern — Terse highlights it and logs it for the CLAUDE.md rule generator.

Session Analysis
📋

CLAUDE.md Rule Generation

After each session, Terse analyzes patterns and writes CLAUDE.md rules that prevent the same inefficiencies from happening again. Your sessions get cheaper over time.

Unique to Terse
💰

Per-Turn Cost Tracking

See exactly how many tokens each turn costs, in real time. Know instantly when a single prompt is burning $0.50 and redirect before the session gets expensive.

Live Tracking

The Claude Code Token Problem

Most of your token spend isn't your prompts — it's the invisible overhead Claude Code generates in every turn.

Without Terse — 2 hour session
CLI output noise~150,000 tokens
Duplicate file reads~30,000 tokens
Verbose prompts~30,000 tokens
Total~210,000 tokens
~$3.15 with Opus · $0.63 with Sonnet
With Terse — same session
CLI output noise~16,500 tokens
Duplicate file reads~3,000 tokens
Compressed prompts~9,000–18,000 tokens
Total~23,000–37,000 tokens
~$0.35–$0.55 with Opus · $0.07–$0.11 with Sonnet

Frequently Asked Questions

Everything about using Terse with Claude Code.

How does Terse reduce Claude Code token costs?
Terse auto-detects Claude Code via process scanning and attaches a floating optimizer. It compresses every prompt (removing filler, typos, redundancy), detects duplicate tool calls, flags redundant file reads, and tracks per-turn costs in real time. In a typical 2-hour session, total token usage drops by 40–89%.
Does Terse work with Claude Code automatically?
Yes. Terse scans running processes every 5 seconds and auto-detects Claude Code as soon as it launches. The floating monitor appears above your terminal with no setup required.
What is CLAUDE.md rule generation?
After monitoring a session, Terse analyzes patterns — repeated tool calls, common prompt structures, redundant context — and generates CLAUDE.md rules. Placing these in your project trains Claude Code to behave more efficiently from the start of every future session.
Does Terse send my Claude Code prompts anywhere?
No. Terse is 100% on-device. It reads text via macOS Accessibility APIs, compresses locally with a Rust engine, and writes back to the terminal. Your code and prompts never leave your machine.
What is a duplicate tool call?
A duplicate tool call happens when Claude Code reads the same file multiple times, or runs the same search twice. Each costs tokens for the call plus the result in context. Terse's Agent Monitor flags these in real time so you can interrupt before costs accumulate.
How much does a Claude Code session cost without Terse?
A typical 2-hour Claude Opus session generates ~210,000 tokens of CLI noise. At $15–$75/M tokens, that's $3–$15 per session. Heavy users spend $200–500/month. Terse compresses this by 40–89%.

Also Works With

Terse supports every major AI coding tool and browser-based AI assistant.

⌨️ Cursor — AI Code Editor → 💬 ChatGPT — Browser → 🤖 Aider, OpenClaw, Copilot CLI → 🌐 Claude.ai, Gemini — Browser →

Stop Burning Tokens on Noise

30-day free trial. No credit card required until your trial ends. Cancel anytime.

Start Free Trial — macOS Calculate Your Savings