Zed

Cut Your Zed Token Costs by 40–70%

Zed is the fast, Rust-native AI editor. Terse auto-appears when you focus Zed, compresses the prompts you type into its AI assistant and agent panel, and tracks per-request cost across Claude and GPT — 100% on-device.

Start 30-Day Free Trial See How It Works
40–70%prompt token reduction
Autoappears on Zed focus
Modelaware cost tracking
0data leaves device

How Terse Works with Zed

Zed's agentic editing runs multiple agent threads in parallel against hosted models like Claude Opus and GPT-5.x. Terse reads the instruction you type into the AI panel via Accessibility APIs, compresses it on-device, and shows the cost before you send.

Zed — AI Assistant
Terse Active
📄 db.rs
📄 pool.rs
📄 schema.rs
📄 main.rs
Agent · Claude Opus 4
Terse detected Zed — model: Claude Opus 4. Tracking token cost on this thread.
Hey, could you please possibly go ahead and add connection pooling to the database module for me? I was really hoping you could also make sure that it gracefully handles timeouts and retries when the connection drops. Add connection pooling to the database module. Handle timeouts and retries gracefully on dropped connections. Terse: 71 → 24 tokens · saved 66% · ~$0.0007 saved · Normal mode
Terse compressed your instruction — removed filler & hedging, priced against Claude Opus 4
−66%
Ask the agent…
👁️

Auto Zed Detection

When you click into Zed's AI assistant or agent panel, Terse auto-appears at the top of your screen — without stealing focus or interrupting Zed's 120fps flow.

Auto-detect
✂️

Prompt & Instruction Compression

The verbose instructions you type into Zed's agent — "could you please go ahead and…", "I was hoping you could…" — get stripped of filler and hedging while intent is preserved.

Normal Mode
🧵

Agent-Thread Context Awareness

Zed runs multiple agent threads in parallel. Terse follows the thread you're typing in, so each thread's prompt is compressed and priced independently.

Per-Thread
💰

Model-Aware Cost Tracking

Terse knows which model your thread uses — Claude Opus, Sonnet, or GPT-5.x — and prices each prompt with that model's real per-million rate so savings are accurate.

Claude · GPT
📊

Session Cost Summary

See tokens and dollars saved across a whole Zed session. On Pro your $5/mo of hosted-model token credits stretch further when prompts stay lean.

Per-Session
🔒

100% On-Device

Terse only reads the text you type, and compresses it locally via Accessibility APIs. It never reads Zed's internal traffic, your code, or your project — nothing leaves your machine.

Private

Why Zed AI Requests Cost Tokens

Zed's cost isn't just your instruction — it's everything the agent sends to the hosted model before your words reach it. On Pro, that spends your $5/mo token credits and then bills by usage.

Token breakdown — Zed agent request
Zed system prompt~1,500 tokens
File & buffer context~6,000–14,000 tokens
Thread history~2,000–5,000 tokens
Your instruction (verbose)~150–700 tokens
Your instruction is the only part you control. Terse cuts it by 40–70%.
What Terse does
→Removes filler phrases ("could you please go ahead", "I was hoping")
→Fixes typos that cause misinterpretation and re-prompting
→Prices every prompt against the model your Zed thread has selected
→Compresses redundant phrasing while preserving full meaning
Result: leaner prompts, and your Pro token credits go further

New to how agent pricing adds up? Read the full AI coding agent cost guide.

Supported Models

Terse tracks per-request cost with each model's real input pricing, matched to the model your Zed thread has selected — including Zed's hosted models.

Claude Opus 4
$15.00 / M input
Claude Sonnet 4.5
$3.00 / M input
GPT-5.x
$1.25 / M input
GPT-4o
$2.50 / M input
Gemini 2.5 Pro
$1.25 / M input
Claude Haiku 4.5
$1.00 / M input

Zed exposes hosted models on Pro (with $5/mo token credits) or your own API keys. Terse matches whichever model you select. Prices approximate per-million input tokens.

Frequently Asked Questions

Everything about using Terse with Zed.

Does Terse work with Zed?
Yes. Terse reads the text you type into Zed's AI assistant and agent panel via macOS Accessibility APIs (and Windows UI Automation). A floating bar auto-appears when you focus Zed, showing live token counts and the compressed version of your prompt. It works on your typed instructions — not on Zed's internal AI traffic.
Why do Zed AI requests use tokens and credits?
Every agent request sends your instruction plus code context, thread history, and a system prompt to a hosted model. On Zed Pro you get $5/mo of token credits and then pay by usage — so shorter, denser prompts directly stretch your credits. Terse compresses the editable part by 40–70%.
Does Terse read my Zed code?
No. Terse only reads the natural-language prompt you type into the AI panel, and processes it entirely on-device. It never reads your source files, your project, or Zed's internal context, and nothing is sent to a Terse server.
Which models does Terse track?
Claude Opus 4, Claude Sonnet 4.5, Claude Haiku 4.5, GPT-5.x, GPT-4o, and Gemini 2.5 Pro — including Zed's hosted models. Terse is model-aware and prices each prompt with the real per-million rate of whichever model your thread has selected.
Will Terse break my Zed prompts?
No. In Soft and Normal modes, meaning is fully preserved — only filler, hedging, and typos are removed. Code blocks, file paths, and technical terms are always protected. You can preview the compressed version before applying, or use Aggressive mode when you want maximum compression.
Is there a free trial?
Yes — both Pro ($4.99/mo) and Premium ($99/mo) include a 30-day free trial. No credit card charge until your trial ends. Cancel anytime.

Also Works With

Terse supports every major AI code editor and coding assistant.

🖱️Cursor — AI IDE→ 🏄Windsurf — AI IDE→ ⚡Claude Code — CLI Agent→ 🤖GitHub Copilot — VS Code→

Make Every Zed Agent Prompt Leaner

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

Start Free Trial — macOS & Windows Calculate Your Savings