Optify AI — Token Optimizer

Cut your AI inference costs automatically. Zero configuration. Zero data collection. Runs entirely on your machine.

Optify AI compresses your prompts before they reach the model — silently removing filler words, redundant phrasing, and low-signal tokens — so you send fewer tokens without losing meaning. The result: lower costs and faster responses, every session.

How it works

When you chat with @optify-ai in Copilot Chat, Optify AI compresses your message before forwarding it to GitHub Copilot. The response comes back as normal. You never notice a difference in quality — only in the token count.

Compression runs locally using a neural model (LLMLingua-2, MIT licensed, by Microsoft Research) that runs entirely on your CPU. No prompt ever leaves your machine before being sent to Copilot.

Quick start

1. Install the extension
2. Open GitHub Copilot Chat (Ctrl+Shift+I / Cmd+Shift+I)
3. Type @optify-ai followed by your question

@optify-ai how do I reverse a linked list in Python?

That's it. Optify AI handles the rest automatically.

Status bar

The ⚡ Optify AI item in the bottom status bar shows your current session stats (tokens saved, estimated cost reduction). Click it to toggle optimization on or off.

Commands

Compatibility

Currently supported: GitHub Copilot Chat only.

Optify AI works with VS Code's GitHub Copilot Chat (@optify-ai in the chat panel). Support for other providers (Claude, Cursor, direct API) is coming in a future release.

Requirements:

• VS Code 1.90 or later
• GitHub Copilot subscription (any tier)

Privacy — your prompts stay on your machine

Optify AI is built with privacy as the first principle:

• No prompt content is ever stored or transmitted. Compression happens locally before the prompt is sent to Copilot.
• No telemetry. Token counts are tracked in memory for the current session only and cleared when you close VS Code.
• No accounts, no sign-in, no backend calls. The extension runs entirely offline except for the standard Copilot Chat connection you already have.
• No data collection of any kind. We do not collect, transmit, or sell any information about you or your code.

Session stats (tokens saved, estimated savings) are calculated locally and exist only in your current VS Code session.

First-run model download

On first install, Optify AI silently downloads the neural compression model (~178MB) from HuggingFace and saves it to your VS Code extension storage folder. This happens once in the background and requires an internet connection. After that, everything runs fully offline.

You can watch the download progress in the Optify AI output panel (View → Output → Optify AI).

Settings

License

MIT — see LICENSE. Third-party attributions (including LLMLingua-2 by Microsoft Research) are listed in NOTICES.