OMNI

Your agent reads everything your terminal prints, then reads most of it again next turn. OMNI drops the noise before the model sees it, and hands back a reference for the lines it has already shown. Nothing is deleted, and it never invents a result.

🇺🇸 English | 🇯🇵 日本語 | 🇨🇳 简体中文 | 🇸🇦 العربية | 🇮🇩 Bahasa Indonesia | 🇻🇳 Tiếng Việt | 🇰🇷 한국어

brew install fajarhide/tap/omni && omni init

Drops the noise. Build logs, Docker layer hashes, progress bars, ANSI colour. The part of the output nobody reads is removed before it reaches the model.

Stops re-sending what the agent has already seen. A run of lines it was shown earlier in the session comes back as one marker with a handle, not as the bytes again. This is the half a filter cannot do: it removes bytes because they are already in the context, not because a pattern calls them noise.

Remembers across sessions. Restart your editor or switch agents, and the project context is still there.

Gets out of the way. A failing command passes through verbatim. JSON, YAML and CSV are never touched. Most commands are handed back unchanged, and that is the intended behaviour rather than a gap.

Without OMNI, one commit's Author / Date / body already fills the screen. With OMNI, every commit is kept, as one hash subject line. Nothing is summarised away.

Without OMNI raw git log -15	With OMNI every commit kept, 94% smaller

Measured on the fixtures in tests/fixtures/, so you can reproduce any row:

That last row is the point of the table. A pod listing is an enumeration where every row is a datum, so there is nothing to drop, and OMNI reports nothing rather than inventing a saving.

Four guarantees, each one a link to the code or the issue that made it true rather than a sentence asking you to trust it.

Three tiers, already in the schema, never written down until now. The short answer to "will OMNI still know my project after a month away" is yes for the conclusions and no for the raw bytes.

The boundary this sets is worth stating plainly, because it is the one thing a handle cannot promise: omni retrieve for content archived more than 30 days ago will not resolve. Hold the shortest window open while measuring with OMNI_TRACE_RETENTION_DAYS=90.

omni reset wipes all of it, and omni doctor shows the live counts.

omni doctor prints the tier for every installed host. Savings are only ever counted where the model actually received less.

Codex CLI needs one extra step. It runs only hooks it has been told to trust and skips the rest without a word, so after omni init --codex start codex once and approve them under "Hooks need review". omni doctor fails until you do. See #359.

macOS / Linux:

brew install fajarhide/tap/omni
omni init      # interactive setup for Claude, Cursor, VS Code, Codex, Antigravity
omni doctor    # verify, or `omni doctor --fix`

Universal (macOS / Linux / WSL):

curl -fsSL omni.weekndlabs.com/install | bash

Windows (PowerShell):

irm omni.weekndlabs.com/install.ps1 | iex

Then run your commands normally. There is nothing to prefix and no proxy to wrap.

Every figure OMNI publishes states the corpus it came from and the week it covers, because execution_traces is pruned after seven days and a number that outlives its corpus cannot be checked by anyone, us included.

On the 2026-08-03 to 08-10 UTC window, replayed on the release binary over real command executions that reached a model:

• Build and test output: 76.9%. File re-reads, the largest class: 0.0% from the filters and 26.3% from the ledger, which is the gap the ledger exists for.
• 97.3% of calls saved nothing at all, and we publish that because it tells you what the rest are worth. No call came back larger in this measurement. There were 2 until (#398), and we published them while they stood.
• 21 ms per command, growing with your history rather than with the payload. On a 205 MB database it is 61 ms.

Full corpus, per-class breakdown, fixtures and latency tables: docs/BENCHMARKS.md. Reproduce them with cargo test --release --test bench_replay -- --ignored.

Does OMNI permanently delete my logs? No. Raw logs are stored locally in the SQLite RewindStore. The agent receives a hash and can retrieve the full log at any time.

Will this slow down my terminal? Measurably, yes, and the cost grows with your history rather than with the payload. A 496-byte git status takes about 21 ms against a fresh database and 61 ms against a 205 MB one. Budget for it. OMNI_PASSTHROUGH=1 skips the pipeline entirely.

Can I add my own filters? No, and that is deliberate as of 0.7.0. The filters are compiled into the binary, so the set that runs is the set the tests cover and there is no file on disk that changes what your agent is shown. Two tiers were removed to get there: a project's own .omni/signals/, which made a filter a thing a repository could ship to its visitors, and ~/.omni/signals/. The whole filter layer is worth 804 bytes over 6,656 recorded commands, so what it cost in surface it was not paying back. If a tool needs a signal, open an issue and it ships in the binary for everyone.

How do I get back something OMNI folded? omni retrieve <handle>, where the handle is the 16 characters inside the marker. It works on every host, with or without MCP. Agents that have the MCP server wired can call omni_retrieve instead.

Can I watch the numbers instead of running a command? omni dashboard serves them on 127.0.0.1, read-only, from the same database omni stats reads. It binds loopback and nothing else.

How do I see my own savings? omni stats after a few days. It leads with session lifetime, how many commands a session carries before the host closes it, because that is what the context window costs you. The distillation percentage below it is a diagnostic for one host's pipeline, not a product claim. omni stats --share prints a copy-pasteable summary, and omni stats --card writes it as an image.

• How it works, and what it costs: pipeline, RewindStore, the Memory OS
• Benchmarks in full: corpus, per-class, fixtures, latency
• Contributing: make ci and you're in

brew install fajarhide/tap/omni && omni init

A passion project for the era of agentic AI. Contributions welcome. MIT License.

Built with care by Fajar Hidayat