Compresso
50%offYour first 3 monthsv0.1.1 out now

Same answers.
Fewer tokens.

Compresso runs a local AI compression layer between Claude Code or Codex and the provider. It shrinks and compresses what your agent reads, tool output, logs, files, history, before the request is billed.

Download for macOSSee how it works5-day free trial · no card up front
32,674,025
tokens saved
$98
saved · at list input price

The app

Everything you need, in one screen.

Live status you can pause at any moment. Savings split by agent and by day, over seven or thirty. And a connect switch per agent that puts your config back exactly as it was.

The Compresso overview screen: cost saved, cache discounts and context optimized across the top, above a seven-day chart splitting Claude Code and Codex into tokens sent and tokens removed.
  • Manage all your agents in one place
  • Per-agent, per-day usage/savings breakdown
  • Connect and disconnect without editing files

Check all the app features →

How it works

One local hop, tons of savings!

  1. Your prompt

    Tool output, logs, files, history.

  2. On your Mac

    CacheAligner → ContentRouter

  3. Anthropic · OpenAI

    Under your own credentials.

How it works, in detail →

Proof

Real token savings, per use case.

Measured on whole agent tasks, not on a hand-picked payload.

Savings
Code search
17,7651,408
92%
SRE incident debugging
65,6945,118
92%
GitHub issue triage
54,17414,761
73%
Codebase exploration
78,50241,254
47%

What survives compression?

98.2%

of the original content is still there, at 94.9% compression.

How much does it save?

up to 92%

on the heaviest of the workloads above. The lightest still saves 47%.

What does it cost?

52ms

added to the median request, measured in production.

See the benchmarks in detail →

Pricing

Free for 5 days. Then a tier that matches your plan.

5 days

free trial

50% off your first 3 months, from $1.67/mo. No card to start. When the trial ends, subscribe from inside the app, or don’t, and Compresso quietly steps out of the way. Up to 2 Macs per account.

See pricing

What it never does

We take security seriously.

Compresso is a compression layer built on open source engines, starting with headroom (Apache-2.0), so you can read exactly what runs on your Mac. More engines are on the way.

Never collects your keys

Your Authorization header is copied byte for byte, never stored, never logged, read on your Mac only to name your plan. For Claude, we ask Anthropic.

Never sees your prompts

Optimization is local. Your code and prompts go to Anthropic or OpenAI directly, never through our servers.

Never opens a port

Loopback only, with requests from other origins rejected outright.

Never fingerprints

A random device ID in your Keychain. No hardware identifiers, no third-party analytics.

Never edits your shell

No rc files, no PATH changes, no hooks injected into your tools.

Leaves nothing behind

Uninstall restores every config it changed and removes everything it wrote, listed up front before it runs.

How each boundary is enforced →

Try it on today’s session.

Enjoy fewer tokens from your first request.
5 days free, no card.

Download for macOSApple Silicon · macOS 14+