Skip to content
| Marketplace
Sign in
Visual Studio Code>Machine Learning>Terse — AI Agent Monitor & Token OptimizerNew to Visual Studio Code? Get it now.
Terse — AI Agent Monitor & Token Optimizer

Terse — AI Agent Monitor & Token Optimizer

Lucas Zeng

|
1 install
| (0) | Free
Monitor Claude Code, Copilot & Cursor live. Token counts, cost, cache efficiency, burn rate per session. Compress prompts 20–40% with 20+ techniques. Auto-routing suggestions.
Installation
Launch VS Code Quick Open (Ctrl+P), paste the following command, and press enter.
Copied to clipboard
More Info

Terse — Token Optimizer & Agent Monitor for VS Code

Compress AI prompts by 20–40% and monitor every AI agent running in your editor in real time.

Terse brings the full macOS app experience into VS Code: live session monitoring for Claude Code, Copilot Chat, Cursor, Continue, Cline, and Aider — plus a prompt optimizer with 20+ compression techniques, spellcheck, and smart model routing suggestions.


⚡ Agent Monitor

Terse automatically detects AI agents running in your VS Code environment and shows live session stats — no setup required.

Supported Agents

Agent Detection Session Data
Claude Code Process scan Live JSONL streaming · exact token counts · cache efficiency
GitHub Copilot Chat Extension API Detected · activity indicator
Cursor Agent App name / process Conversation history · estimated tokens
Continue.dev Extension API Detected · activity indicator
Cline Extension API Detected · activity indicator
Roo Code Extension API Detected · activity indicator
Aider Process + .aider.chat.history.md Conversation turns · estimated tokens

Live Stats per Agent

  • Context fill meter — visual bar showing % of context window used (warning at 60%, danger at 85%)
  • Token count — exact (Claude Code) or estimated (others)
  • Cost in USD — real-time cost calculated from Anthropic API pricing
  • Burn rate — tokens/min based on rolling 5-minute window
  • Cache efficiency — % of input tokens served from prompt cache (Claude Code only)
  • Activity feed — last 5 messages with role labels (user / assistant / tool)

Auto-Routing Suggestions

Terse analyzes your session and recommends a cheaper model when appropriate:

  • Short/simple sessions → "Route to Haiku for 80% cost savings ($0.80/M vs $3/M)"
  • Long context → "Context 85% full — compress or start a new session"
  • Low cache hit rate → "Add a CLAUDE.md to improve prompt caching"
  • Redundant tool calls → "package.json read 3× this session"
  • Large tool results → "3 large tool results — consider truncating output"

Claude Code Hooks

For Claude Code specifically, Terse can install PostToolUse hooks that compress tool outputs before they enter the context window — reducing bloat from large file reads, grep results, and terminal output.

Click Install Hooks on the Claude Code card in the Agent Monitor panel.


🧠 Prompt Optimizer

3 Optimization Modes

Mode What it does
Soft Contractions + whitespace only. 100% meaning-preserving.
Normal Removes filler, hedging, politeness, meta-language. Balanced.
Aggressive Everything in Normal + abbreviations + telegraph-style. Max savings.

20+ Optimization Techniques

  • Filler & hedging removal (basically, I think that, perhaps)
  • Phrase shortening (in order to → to, make a decision → decide)
  • Meta-language removal (I want you to, the following is, let me explain)
  • Question-to-imperative (Can you explain X? → Explain X.)
  • Redundant modifier collapse (clear and concise → concise)
  • Politeness stripping (greetings, apologies, thank-yous in 8 languages)
  • Vocabulary simplification (utilize → use, facilitate → help)
  • Passive voice removal
  • Self-context removal (I'm a developer working on… → removed)
  • Log & terminal output deduplication
  • Stack trace compression (first + last frame only)
  • Code signature extraction (declarations only, no bodies)
  • Contraction expansion (do not → don't)
  • Numeralization (twenty → 20)

Spellcheck Pipeline

Every optimization runs through a 3-layer spellcheck:

  1. 400+ hardcoded coding-domain typos (fucntion → function, databse → database)
  2. Hunspell English dictionary via nspell
  3. Context-aware real-word error correction

Quick Start

  1. Install the extension
  2. Sign in via the Terse sidebar (activity bar lightning bolt icon)
  3. Agent monitoring starts automatically — open the sidebar to see detected agents

For prompt optimization:

  • Select text → right-click → Terse: Optimize & Replace
  • Or press Cmd+Shift+O (Mac) / Ctrl+Shift+O (Windows/Linux)

Keyboard Shortcuts

Shortcut Action
Cmd+Shift+T Open Terse panel
Cmd+Shift+O Optimize & replace selection
Cmd+Shift+Y Copy optimized text to clipboard

Commands

All commands available in the Command Palette (Cmd+Shift+P):

  • Terse: Show Panel
  • Terse: Optimize Selection
  • Terse: Optimize & Replace
  • Terse: Copy Optimized
  • Terse: Set Optimization Mode
  • Terse: Toggle Auto Mode
  • Terse: Sign In
  • Terse: Upgrade Plan

Settings

Setting Default Description
terse.mode normal Optimization mode: soft, normal, aggressive
terse.autoMode false Auto-optimize selection as you type (debounced)
terse.autoReplace false Auto-replace selection when auto-mode fires
terse.showStatusBar true Show token count in status bar
terse.debounceMs 600 Debounce delay for auto-mode (ms)

Plans

Plan Features
Free Trial (30 days) Full access to all features, all agents
Pro Unlimited optimizations, up to 3 devices
Premium Unlimited everything, priority support

Sign in from the sidebar panel. No credit card required for trial.
Your Terse account works across the macOS app and VS Code extension.


Privacy

  • Text is optimized locally in VS Code — your prompts never leave your machine.
  • Agent monitoring reads local files only (~/.claude/projects/, Cursor SQLite, etc.).
  • Auth and license checks use HTTPS to terseai.org.
  • No telemetry, no prompt logging.
  • Contact us
  • Jobs
  • Privacy
  • Manage cookies
  • Terms of use
  • Trademarks
© 2026 Microsoft