omni

Less noise. More signal. Right signal. Reduce AI token consumption by up to 90%.

The Semantic Signal Engine that cuts AI token consumption by up to 90%.

OMNI acts as a context-aware terminal interceptor—distilling noisy command outputs in real-time into high-density intelligence, ensuring your LLM agents work with meaning, not text waste.

Why OMNI?

AI agents are drowning in noisy CLI output. A git diff can easily eat 10K tokens, while a cargo test might dump 25K tokens of redundant noise. Claude and other agents read all of it, but 90% of that data is pure distraction that dilutes reasoning and drains your token budget.

OMNI intercepts terminal output automatically, keeping only what matters for your current task. It’s not just about making output smaller; it’s about making it smarter. By understanding command structures and your active session context, OMNI ensures your agent sees the signal, not the waste.

1. Cost & Latency: Large outputs consume your context window rapidly and increase the cost of every message.
2. Cognitive Dilution: LLMs can lose track of complex reasoning when buried under megabytes of raw CLI logs.
3. Auto-Truncation: Claude Code often cuts off large outputs, potentially missing the exact error or diff line it needs to see.

OMNI is the solution. It acts as a "Sieve" that sits between your terminal and the AI, turning raw data into Semantic Signals.

How It Works: The Signal Lifecycle

OMNI employs a unique, multi-layered native interception strategy to ensure maximum efficiency without losing information:

1. Surgical Pre-Hook (PreToolUse)

Intercepts noisy commands (like git, cargo, npm, pytest) before they execute. By natively rewriting these commands to omni exec, OMNI prevents auto-truncation and ensures the AI sees a distilled, high-density stream from the first line.

2. Safety-Net Post-Hook (PostToolUse)

Automatically distills output from any tool after it runs. This acts as a backup for custom scripts or unknown commands.

Real-time ROI feedback on every distilled command.

3. Session Continuity (SessionStart)

When you start a new Claude session, OMNI injects a high-level summary of your previous state—hot files, last errors, and active task context—so the agent never reaches for "context" it already had.

4. Smart Compaction (PreCompact)

Before Claude prunes its conversation history to save space, OMNI provides a permanent summary of the work done so far, ensuring long-term project memory stays sharp.

The Impact

Reduce AI Token Usage by up to 90%
Zero Information Loss. Native Binary Performance. Real-time ROI Monitoring.

Key Features

RewindStore: Zero Information Loss

When OMNI distills output, the original raw content isn't discarded—it's archived in the RewindStore with a SHA-256 hash.

• Agent Access: Call omni_retrieve("hash") via the MCP tool.
• Human Access: Use omni rewind list and omni rewind show <hash> to manage your archives locally.

Signal Comparison: omni diff

Instantly visualize the value of OMNI. Run omni diff after any command to see a side-by-side comparison of the raw input vs. distilled signal.

Session Intelligence

OMNI doesn't just compress; it understands context. It tracks which files you are editing ("Hot Files") and which errors are recurring.

Transcript & Recovery

OMNI safely persists session transcripts as you work. If your AI agent crashes or gets interrupted, you can seamlessly resume your session and pick up right where you left off. OMNI ensures you never lose critical tool calls or outputs. Use omni session --resume to recover an interrupted session.

Pattern Discovery (Learning)

OMNI automatically collects samples of repetitive noise in the background. Use omni learn --status to discover new candidate filters.

The OMNI Philosophy: Deliberate Action

OMNI is designed for maximum safety and control. By default, core commands like init, session, and learn will only show a help screen if no flags are provided. This prevents accidental changes to your global configuration.

Every core command follows a consistent Discovery vs. Action pattern:

• Discovery: Use --status to see what OMNI has found (installation status, session details, or new noise patterns).
• Action: Use explicit flags like --all, --apply, or --clear to commit changes.

Analytics Dashboard

Keep track of your project's efficiency with OMNI's built-in reporting:

Quick Start

# 1. Install via Homebrew (macOS/Linux)
brew install fajarhide/tap/omni

# 2. Perform Full Setup (Hooks + MCP Server)
omni init --all

# 3. Verify Installation
omni doctor

# 4. Or auto-fix any issues
omni doctor --fix

# 5. Check Current Status
omni init --status

On universal setup

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

Custom Filters (TOML)

You can define your own distillation rules for custom internal tools:

# ~/.omni/filters/deploy.toml
schema_version = 1

[filters.deploy]
description = "Internal deployment tool"
match_command = "^deploy\\b"

[[filters.deploy.match_output]]
pattern = "Deployment successful"
message = "deploy: ✓ success"

strip_lines_matching = ["^\\[DEBUG\\]", "^Connecting"]
max_lines = 30

Architecture

flowchart TB
    Agent["Claude Code / MCP Agent"]
    
    subgraph Hooks["Native Hook Layer (Transparent)"]
        Pre["Pre-Hook\n(Rewriter)"]
        Post["Post-Hook\n(Distiller)"]
        Sess["Session-Start\n(Context)"]
        Comp["Pre-Compact\n(Summary)"]
    end

    Agent --> Pre
    Pre -->|"omni exec"| Output["Raw Stream"]
    Output --> Post
    Post --> Agent

    subgraph OMNI_Engine["OMNI — Semantic Signal Engine"]
        direction LR
        C["Classifier"]
        S["Scorer\n(Context Boost)"]
        R["Composer\n(Signal Tiering)"]
        C --> S --> R
    end

    Post --> OMNI_Engine
    Pre --> OMNI_Engine
    
    subgraph Persistence["Persistence Store (SQLite)"]
        ST["SessionState"]
        RW["RewindStore"]
    end

    OMNI_Engine <--> Persistence
    Sess --> ST
    Comp --> ST

Development

OMNI is built for high-performance AI workflows with professional standards.

make ci              # Run fmt, clippy, tests, and security audit
cargo build          # Build the binary
cargo test           # Run all 147 tests
cargo insta review   # Review and accept snapshot changes

See docs/TESTING.md for a detailed breakdown of our 190+ test suite covering Context Safety, E2E Hooks, Security, and Performance Assertions.

See CLAUDE.md, CONTRIBUTING.md, and Critical Guardrails for the full contributor guide and architectural rules.

Star History

License

MIT © Fajar Hidayat