Prompt Router
Prompt Router automatically detects whether your GitHub Copilot prompt needs Ask, Plan, or Agent mode — and routes it there instantly, saving you tokens and context-window budget with zero extra clicks.
Why Prompt Router?
Every Copilot chat prompt costs tokens. Agent mode is powerful but expensive — it reads files, runs terminals, and burns through your context window even for simple questions. Prompt Router silently classifies each prompt and steers you toward the cheapest mode that can actually answer it.
Think of it like a smart receptionist: it reads your note and says "this goes to the quick-answer desk, not the full-orchestration team."
Features
| Feature |
How to use |
| Auto-classify any prompt |
Press Ctrl+Shift+R (Mac: Cmd+Shift+R), type your prompt, and Copilot opens in the right mode automatically |
@smart participant |
Type @smart followed by your prompt — see the detected type label + a one-click mode-switch button |
@router participant |
Type @router for a full score breakdown (Ask / Plan / Agent percentages) |
| Smart dropdown agent |
Select Smart from the chat mode dropdown for token-efficient auto-routing in the conversation itself |
| Token savings dashboard |
Run Prompt Router: Show Token Savings Dashboard from the Command Palette |
| Learns your patterns |
Accepts/dismissals train a lightweight local model that personalises routing over time |
How It Works (Plain English)
- You type a prompt.
- Prompt Router runs it through a heuristic classifier (keyword signals) blended 40/60 with a Naive Bayes ML model.
- Learned pattern boosts from your past corrections are applied on top.
- The highest-scoring mode wins.
- VS Code chat opens in that mode with your prompt pre-filled — no copy-paste needed.
The classification runs entirely locally — no network calls, no telemetry.
Mode Guide
| Mode |
When to use |
Token cost |
| Ask |
Explanations, definitions, debugging concepts, what/why/how questions |
~500 tokens |
| Plan |
Architecture, design, step-by-step approach, best practices |
~800 tokens |
| Agent |
Write, fix, create, refactor, run terminal commands, file operations |
~3000+ tokens |
Quick Start
- Get a License — one-time purchase, use on up to 3 devices.
- Install the extension and activate with your license key.
- Press
Ctrl+Shift+R and type any Copilot prompt.
- Watch it open in the right mode automatically.
Or use @smart in any existing chat session for a type label + one-click switch button.
Buy License — $2.9 / ₹19 · One-time payment · 3 device activations · Instant email delivery
Commands
| Command |
Shortcut |
| Ask Copilot (Auto Mode) |
Ctrl+Shift+R |
| Show Token Savings Dashboard |
Command Palette |
| Clear Learned Patterns |
Command Palette |
Settings
| Setting |
Default |
Description |
promptRouter.confidenceThreshold |
0.45 |
Minimum confidence before showing a suggestion |
promptRouter.autoDismissSeconds |
8 |
Seconds before status bar tip auto-dismisses |
promptRouter.enabled |
true |
Enable or disable the extension entirely |
Requirements
- VS Code
^1.99.0
- GitHub Copilot Chat extension installed and signed in
Release Notes
0.1.0
- Initial release
- Heuristic + ML hybrid classification
@smart and @router chat participantsCtrl+Shift+R quick-input shortcut- Token savings dashboard
- Pattern memory (local learning)
Pricing
One-time purchase — no subscriptions, no recurring fees.
|
|
| Price |
$2.9 / ₹19 |
| Activations |
Up to 3 devices |
| Delivery |
Instant license key via email |
| Payment |
Razorpay (India) · PayPal (International) |
Buy Now →
License
MIT
