Edgee Claude Code Compressor V2

Token compression for coding agents isn't a sexy problem—but it's a billion-dollar one. Edgee just dropped Compression V2 and they're claiming 50% token reduction on typical coding sessions without losing a single semantically important detail. That's 18,420 tokens trimmed down to 9,210. On a tool that processes 99% of its token volume in the input layer (tool results, tool definitions, codebase context), that's not incremental—that's architectural.

Here's how it works: three techniques across two layers. Layer 1 hits the input—tool result trimming strips 10%, tool surface reduction (new in V2) drops another 10%. Layer 2 attacks the output with output brevity, slicing 30% off the response. All three toggles are independent config flags, and Edgee claims it's 100% semantically lossless on code tasks. The gateway overhead? Under 12ms P50. They're calling it "semantically lossless"—which is a bold claim, but the numbers speak for themselves.

The kicker: it's a drop-in CLI wrapper. Zero code changes. Works with your existing API keys, your existing Anthropic and OpenAI setups. You're not locked into a new platform—you just point Edgee's gateway at your existing workflow and watch the bill drop. For teams running Claude Code, Codex, or any coding agent at scale, this is pure margin capture. Every token you don't send is money you don't spend.

Edgee is building the infrastructure layer for anyone shipping AI coding agents. If you're burning through context windows and API credits, they're offering a shortcut that doesn't require rewriting your stack.