Edgee Compresses Your Agent's Tokens at the Gateway
Your coding agent burns through tokens like they're free. They're not. Edgee is an open-source Rust gateway that sits between your agent and the LLM provider, compressing tool call results before they reach the model. File listings, git logs, build output, test results β all the noise that inflates your context without adding information gets squeezed out.
The compression is lossless from the model's perspective. Responses come back identical, but prompts are leaner. For Claude Code API users, that's up to 30% cost reduction. For Pro and Max subscribers, up to 35% longer sessions before you hit the ceiling. Works with Codex, Cursor, and any OpenAI-compatible client too.
The setup is minimal: drop Edgee in front of your existing workflow, no code changes needed. It also handles multi-provider routing and real-time usage tracking so you can see exactly where your tokens go. The whole thing is Apache 2.0.
Only 41 GitHub stars so far, which tells you it's early. But the premise is solid and it just hit the top 5 on Product Hunt with 131 upvotes. Agents generate enormous tool output that's mostly redundant, and squeezing it before it enters the context window is an obvious win that nobody else is doing at the gateway level.
https://github.com/edgee-ai/edgee
← Back to all articles
The compression is lossless from the model's perspective. Responses come back identical, but prompts are leaner. For Claude Code API users, that's up to 30% cost reduction. For Pro and Max subscribers, up to 35% longer sessions before you hit the ceiling. Works with Codex, Cursor, and any OpenAI-compatible client too.
The setup is minimal: drop Edgee in front of your existing workflow, no code changes needed. It also handles multi-provider routing and real-time usage tracking so you can see exactly where your tokens go. The whole thing is Apache 2.0.
Only 41 GitHub stars so far, which tells you it's early. But the premise is solid and it just hit the top 5 on Product Hunt with 131 upvotes. Agents generate enormous tool output that's mostly redundant, and squeezing it before it enters the context window is an obvious win that nobody else is doing at the gateway level.
https://github.com/edgee-ai/edgee
Comments