This Indian engineer working at Netflix just built an open source tool that cuts your AI token usage by 60 to 95%.

@ai.with.andrew

48.4K followers · usually 2.3K views a post

Views

99.7K

Likes

3.6K

Comments

29

Shares

811

SoundTikTok sound · original sound - ai.with.andrew

Promoted product

Headroom

Compress tool outputs, logs, files, and RAG chunks before they reach the LLM. 20% fewer tokens for coding agents, 60-95% fewer tokens for JSON, same answers. Library, proxy, MCP server. - headrooml...

3 videos3 creators

Why it worked

The video leverages a strong 'hero' narrative by highlighting a specific developer's achievement and provides immediate, quantifiable value (95% savings) to a technical audience.

Summary

The video introduces an open-source tool called Headroom that significantly reduces AI token usage for developers. It explains how the tool functions as a compression layer and provides instructions on how to set it up.

Structure

1Introduce the tool and its primary benefit
2Explain the technical mechanism of token compression
3Demonstrate the simple setup process
4

Original caption

send a DM with Headroom savings Tejas Chopra just launched Headroom, an open-source tool that slashes AI token usage by up to 95% without losing output quality. Perfect for developers using cloud code, Cursor, or Copilot looking for a smart, trending token compression solution. #headroomtool #tejashchopra #aitokenoptimization #opensourceai #cloudcode

More from @ai.with.andrew

Product placement

Headroom is the central subject of the video, shown via screen recordings of its GitHub repository, terminal commands, and performance charts.

Call to action

send a DM with Headroom savings

On-screen text

this indian engineer Token usage Tejas Chopra headroom Compressed setup is simple 30,000 stars

Transcript

His name is Tejas Chopra and the tool is called Headroom. It sits between your app and the model, so every tool output, log, and chat history passes through it and gets compressed before it reaches the model. No summarizing and loss, just the same answer with way fewer tokens. Setup is simple. One pip install, then run headroom dash wrap dash cloud, or started as a proxy with zero code changes, and you instantly see the savings. It works with Claude Code, Cursor, and Copilot, and crossed almost 30,000 stars on GitHub.