<div align="center" <img src="media/logo.png" alt="OMNI Logo" width="300" /
<h1OMNI</h1 <p align="center" <emNoise-canceling context and long-term memory for your AI agent — <blossy, but always reversible, and it never fabricates a result.</b Stop paying Claude to read 10,000 lines of terminal noise.</em </p
🇺🇸 English | 🇯🇵 日本語 | 🇨🇳 简体中文 | 🇸🇦 العربية | 🇮🇩 Bahasa Indonesia | 🇻🇳 Tiếng Việt | 🇰🇷 한국어
</br</br <b 58.9% fewer tokens on a real command mix · Cross-Session Memory · Format-safe · Always reversible · Fails open, never fabricates · Numbers you can reproduce </b
</br</br <img src="media/demo.gif" alt="OMNI distilling a noisy cargo test run down to the verdict, then omni stats" width="820" / </div
---
Every AI coding assistant has two massive problems.
1. They read everything. Build logs. Docker logs. CI logs. Progress bars. ANSI colors. Thousands of tokens... to find one line. Claude isn't expensive. Your terminal is.
2. They forget everything. Every time you restart Cursor, or switch from Claude Code to Windsurf, your agent gets amnesia. You have to re-explain the project goal. You have to remind them of the same framework gotchas over and over again.
OMNI fixes both.
---
The Difference
Problem 1: Your terminal drowns out the signal
The same git log side by side. Without OMNI, one commit's Author / Date / body already fills the screen. With OMNI, every commit is kept — as one hash subject line, 94% smaller. Nothing is summarised away; the footer is measured from the real byte counts, not promised.
<table <tr <td align="center"<bWithout OMNI</b<br/<subraw <codegit log -15</code</sub</td <td align="center"<bWith OMNI</b<br/<subevery commit kept, 94% smaller</sub</td </tr <tr <td valign="top"<img src="media/demo-git-without.gif" alt="a raw verbose git log -15 — one commit's Author, Date and body fill the screen" width="400" /</td <td valign="top"<img src="media/demo-git-with.gif" alt="the same git log -15 through OMNI: every commit as a compact hash + subject line, 94% smaller" width="400" /</td </tr </table
Real numbers, measured on tests/fixtures/ and replayed traces — not aspirations:
| Command | Without OMNI | With OMNI | Saved | |---|---|---|---| | cargo test (490 passed, 10 failed) | 16.5 KB of per-test output | the runner's own pass/fail summary | 93% | | kubectl get pods (35 pods, 5 crashing) | the full table | 35 pods \| 30 running, 5 error + the 5 failing pods named | — | | git diff (multi-file) | lockfiles, whitespace, generated churn | the code that actually changed | 45% | | docker build (heavy cache noise) | 9.2 KB of layer hashes and progress bars | the build result, cache hits folded | 37% |
The honest caveat: OMNI compresses noisy successful output. A command that fails is passed through verbatim — a hidden error is worse than an uncompressed one — and structured output (JSON/YAML/CSV) is never touched. It earns its keep on repetitive tool chatter and gets out of the way everywhere else.