The Semantic Core for Agentic AI

The world's first Semantic Density engine for Agentic AI
Eliminating 30–90% of token noise with Zero Semantic Loss.
Transforming chaotic tool output into pure, high-density signal · Powered by Zig + Wasm.

Why OMNI

Supported AI Agents

OMNI integrates seamlessly with popular AI coding assistants:

Agent	Command	Filters
Claude Code	`omni generate claude-code`	Git, Docker, npm
Codex CLI	`omni generate codex`	25 polyglot filters
Antigravity	`omni generate antigravity`	Google Cloud, k8s
OpenCode AI	`omni generate opencode`	67 AI coding filters

AI agents running on Model Context Protocol (MCP) are limited by the quality of the signal they receive. When Claude runs git diff, docker build, or npm install, it is often flooded with "noise"—redundant lines that dilute its reasoning capacity and bloat your context window.

OMNI is the Semantic Core. It sits between your agent and its tools, refining chaotic streams into high-density intelligence. Our goal isn't just to send fewer tokens, but to ensure every token sent is high-signal.

Zero Semantic Loss — We don't just truncate; we distill. Your AI gets the full context, without the fluff.
30% – 90% Token Efficiency — Achieve massive context savings while improving reasoning signal.
Semantic Confidence Scoring — Every token is analyzed and routed: Keep, Compress (Summarize), or Drop.
Cleaner Signal, Better Reasoning — Benchmarks prove LLMs perform better with 50 pure tokens than 500 noisy ones.
< 1ms Engine Latency — Zero-overhead distillation powered by Zig 0.15.2.
Trust Boundary — Military-grade security filters with SHA-256 verification.

CLI Subcommands: Unified Intelligence

OMNI provides a powerful, multi-purpose CLI that consolidates all diagnostic and reporting tools:

Subcommand	Purpose
`distill`	The core semantic engine (default behavior via stdin).
`density`	Analyzes context gain and "Information per Token" metrics.
`monitor`	Unified dashboard for system status, savings trend, and opportunity scanner.
`bench`	High-speed benchmark for semantic throughput.
`generate`	Outputs templates for Claude Code, Codex, Antigravity, OpenCode.
`setup`	Interactive guide for integration and standard aliasing.
`update`	Check for the latest version from GitHub Releases.
`uninstall`	Remove OMNI and clean up all MCP configurations.

How OMNI Works

OMNI sits between your AI agent and the outside world — silently distilling chaotic output into pure, high-density signal.

graph TD
    subgraph Output ["Your Tool Output (Noisy)"]
        A["git diff / status<br/>docker build / logs<br/>kubectl get pods<br/>aws ec2 describe<br/>terraform plan<br/>npm install / audit<br/>etc"]
    end

    subgraph OMNI ["OMNI MCP SERVER"]
        direction TB
        B["LRU Cache<br/>&lt; 1ms hit"]
        C["Filter Engine (Zig + Wasm)<br/>Semantic Distillation"]
        D["Pure Signal<br/>Refined Context<br/>(30–90% token reduction)"]
        E["Metrics & Density<br/>(Performance Report)"]
        
        B --> C
        C --> D
        D --> E
    end

    A -->|"stdin pipe"| B
    E -->|"Pure Signal<br/>Refined Context"| F["AI Agent Platform (Claude/Codex/Antigravity/OpenCode)<br/>Zero Noise reasoning"]

    %% Theme-agnostic Professional Styling
    style OMNI fill:#1d2b3a,stroke:#334155,stroke-width:2px,color:#f8fafc
    style Output fill:#0f172a,stroke:#1e293b,stroke-width:2px,color:#f8fafc
    style A fill:#3b82f6,stroke:#60a5fa,color:#fff
    style B fill:#8b5cf6,stroke:#a78bfa,color:#fff
    style C fill:#06b6d4,stroke:#22d3ee,color:#fff
    style D fill:#10b981,stroke:#34d399,color:#fff
    style E fill:#f59e0b,stroke:#fbbf24,color:#fff
    style F fill:#f59e0b,stroke:#fbbf24,color:#fff

    %% Link Styling (Arrows)
    linkStyle default stroke:#94a3b8,stroke-width:2px

No filter match → passthrough unchanged (zero overhead)

The OMNI Effect

Before OMNI (LLM sees 600+ tokens of noise):

$ docker build .
Step 1/15 : FROM node:18
 ---> 4567f123
Step 2/15 : RUN npm install
... (500 lines of noise) ...
Successfully built 1234abcd

After OMNI Distillation (LLM sees 15 tokens of pure signal):

Step 1/15 : FROM node:18
Step 2/15 : RUN npm install (CACHED)
Step 3/15 : COPY . .
Successfully built!

That's 98% fewer tokens. The LLM gets the same signal — all builds pass — without the noise.

Integration: Using OMNI Everywhere

OMNI is a standard Model Context Protocol (MCP) server.

Claude Code & Claude CLI

The OMNI CLI is for humans, but omni-mcp is for your AI. It allows Claude or Antigravity to use OMNI's distillation tools automatically.

To register OMNI as an MCP server for Claude Code automatically, run:

omni generate claude-code

This command will automatically detect your absolute home path, register OMNI with Claude Code, and merge the polyglot coding filter bundle into your global ~/.omni/omni_config.json.

Verify with:

claude mcp list

Codex CLI

To register OMNI as an MCP server for Codex automatically, run:

omni generate codex

This command uses the local Codex CLI integration flow, registers OMNI with codex mcp add, and automatically merges the codex-polyglot filter bundle into your global ~/.omni/omni_config.json.

Verify with:

codex mcp list

2. Antigravity (Google)

omni generate antigravity

This command will automatically locate your ~/.gemini/antigravity/mcp_config.json, safely merge OMNI's configurations into your existing servers without overwriting them, save the file, and add cloud-native filters for Kubernetes, Terraform, and Docker layers into your global ~/.omni/omni_config.json.

3. OpenCode AI

omni generate opencode

This command:

Registers OMNI as an MCP server in ~/.config/opencode/opencode.json
Installs 67 AI Coding filters for maximum token efficiency
Supports: npm, yarn, pnpm, bun, tsc, eslint, prettier, vite, webpack, next.js, jest, vitest, pytest, ruff, mypy, cargo, go, docker, kubectl, terraform, gradle, and 50+ more tools

Token-Efficient AI Coding Setup:

# Setup OpenCode integration
omni generate opencode

# Then restart OpenCode and verify
opencode mcp list

# Test distillation
git diff | omni
npm run build 2>&1 | omni

Auto-Generate Config

Use the CLI to generate ready-to-paste configurations:

omni generate claude-code    # For Claude Code / Claude CLI
omni generate codex         # For Codex CLI
omni generate antigravity    # For Google Antigravity
omni generate opencode       # For OpenCode AI
omni setup                   # Full interactive guide

"Always-On" Integration: Agent Autopilot

To ensure your AI agent always uses OMNI for maximum token efficiency, add these Custom Instructions to your agent's settings (e.g., in .claudebashrc, Codex workflow notes, or Antigravity custom instructions):

OMNI Directive: You are equipped with the OMNI Semantic Distillation layer. Always prioritize omni_execute, omni_read_file, and omni_list_dir for environment interactions. OMNI is the source of truth for high-density signal. Use it to preserve your context window and speed up your reasoning.

By providing these instructions, the agent will favor OMNI's tools, saving you up to 90% in token costs automatically.

OpenCode AI Integration

OMNI provides 67 specialized filters for OpenCode AI coding workflows, maximizing token efficiency across all development tools.

Quick Setup

omni generate opencode

Supported Tools & Filters

Category	Tools	Example Output
Package Managers	npm, yarn, pnpm, bun, pip, poetry, composer, bundle	`npm: 42 packages added \| 3.2s`
TypeScript/JS	tsc, eslint, prettier, vite, webpack, next.js	`tsc: 0 errors`
Testing	jest, vitest, pytest, cypress, playwright	`jest: 25/25 passed`
Python	ruff, mypy, black, isort	`ruff: all checks passed`
Systems	cargo, go, zig, dotnet	`cargo: finished \| 4.2s`
Containers	docker, docker-compose, kubectl, helm	`docker: built abc123`
Infrastructure	terraform, ansible, packer, skaffold	`terraform: 3 add \| 1 change`
Mobile	gradle, android, flutter, react-native	`gradle: BUILD SUCCESSFUL`
Security	trivy, semgrep, gitleaks, snyk	`trivy: 5 vulnerabilities`

Example Distillation

Before OMNI (600+ tokens):

$ npm install
added 156 packages in 8s
78 packages are looking for funding
  run `npm fund` for details
npm warn deprecated
npm warn invalid
...

After OMNI (15 tokens):

npm: 156 packages added | 8s

Custom OpenCode Instructions

Add to your OpenCode configuration for maximum efficiency:

{
  "instructions": [
    "Always pipe tool output through OMNI for distillation",
    "Use: command | omni for git, docker, npm, pytest outputs"
  ]
}

The Adaptive Intelligence: Proxy & Distillation

OMNI serves as the Intelligent Nerve Center for your development environment, acting as a high-performance wrapper that ensures only high-value information reaches your AI.

1. Zero-Latency Command Proxy (`--`)

Transform any native command into an AI-ready signal instantly. OMNI intercepts the stream and refines it in real-time without adding overhead.

omni -- git status
# Result: Aggregated repository health (30x more dense)

omni -- docker build .
# Result: Cleaned build layers, surfacing only critical transition states.

2. Deep Semantic Distillation (`distill`)

Leverage the OMNI Engine's specialized algorithms to convert chaotic logs into structured intelligence.

Precision Rewrite: OMNI doesn't truncate data; it semantically analyzes the stream to retain "intent-critical" details.
Context Optimization: By compressing 10,000 lines into a 20-line distillation, OMNI effectively expands your AI's reasoning capacity.

3. Ultra-Fast Benchmarking (`bench`)

Prove the efficiency of the OMNI engine:

omni bench 1000

Shows: OMNI processes thousands of requests per second with sub-millisecond latency (< 0.01ms), meaning it adds zero noticeable overhead when used as a proxy.

Available MCP Tools

OMNI exposes high-density tools that replace standard agent context commands:

Tool	Purpose	Token Saving
`omni_list_dir`	Dense, comma-separated directory listing (no JSON overhead).	High
`omni_view_file`	Range-based file reading + Zig distillation.	Massive
`omni_grep_search`	High-density semantic search results.	High
`omni_find_by_name`	Recursive flat file discovery.	Medium
`omni_add_filter`	Add declarative rules without coding.	N/A
`omni_apply_template`	Apply pre-defined bundles (K8s, TF, Node, Codex, Polyglot).	N/A
`omni_execute`	Run ANY command and distill its output.	Massive (30-90%)
`omni_read_file`	Full file distillation (great for logs/SQL/json).	Massive
`omni_density`	Measure gain and reduction metrics.	N/A
`omni_trust`	Trust a project's local `omni_config.json` before loading.	N/A
`omni_trust_hooks`	Verify SHA-256 hashes of hook scripts.	N/A

Easy Filtering: Zero Coding Required

You can extend OMNI's intelligence without touching a single line of Zig.

The agent will use omni_add_filter to update your configuration instantly. It automatically prioritizes your project-local omni_config.json if it exists, otherwise it updates your global ~/.omni/omni_config.json.

2. Apply Technology Templates

Apply bundles of pre-defined rules for your stack via MCP tool:

omni_apply_template(template="terraform")
omni_apply_template(template="codex-advanced") for tsc, eslint, jest, and vitest summaries commonly produced in Codex-driven workflows.
omni_apply_template(template="codex-polyglot") to cover mixed-language Codex loops across JS/TS, Python, Rust, Go, Zig, and pnpm install logs.
omni_apply_template(template="opencode-advanced") for comprehensive AI coding tools (npm, yarn, pnpm, tsc, eslint, jest, vitest, docker, kubectl, and 60+ more)
Language templates: pytest-advanced, ruff-advanced, cargo-test-advanced, pnpm-advanced, zig-advanced, go-test-advanced.
Supported templates: kubernetes, terraform, node-verbose, docker-layers, security-audit, aws-cloud, codex-advanced, pytest-advanced, ruff-advanced, cargo-test-advanced, pnpm-advanced, zig-advanced, go-test-advanced, codex-polyglot, opencode-advanced.

See the DSL_GUIDE.md for full documentation and examples.

Configuration Architecture

OMNI uses a dual-layer, additive configuration system to provide both global consistency and project-specific flexibility.

Layer	Path	Purpose
Global	`~/.omni/omni_config.json`	Your primary rules, shared across all projects and agents.
Local	`./omni_config.json`	Project-specific overrides or additional rules (e.g., custom masking for a specific repo).

How Merging Works

OMNI first loads the Global configuration.
It then loads the Local configuration (if present in your current directory).
The rules are combined. This means rules from both your global setup and your specific project will be applied simultaneously.

Manual Configuration

You can manually edit these files to define rules (exact matching) or dsl_filters (complex semantic logic):

{
  "rules": [
    { "name": "mask_token", "match": "api_key:", "action": "mask" }
  ],
  "dsl_filters": [
    { "name": "my-custom-sig", "pattern": "MY_SIGNAL:", "confidence": 1.0 }
  ]
}

Tip

Use omni generate config to output a complete, well-commented starter template for your configuration.

Lifecycle: Creation & Editing

Event	Action
Installation	The `install.sh` script sets up your global `~/.omni/omni_config.json`.
AI Tooling	Using MCP tools like `omni_add_filter` or `omni_apply_template` will automatically create the file if it doesn't exist.
Manual Edit	You can edit both global and local files manually at any time using any text editor.
AI Proxy	AI agents can dynamically add project-specific rules via the OMNI MCP interface without you leaving the chat.

Security: Trust Boundary

OMNI includes three layers of security to protect your AI agent from malicious inputs:

1. Project Trust Boundary

OMNI will not load project-local omni_config.json until you explicitly trust it. This prevents a cloned repository from injecting malicious filter rules.

How to use:

Clone or open a project that has omni_config.json.
OMNI will log: ⚠ Local config not trusted. Run omni_trust to review and trust.
Call the omni_trust MCP tool — it shows the config contents and SHA-256 hash.
The project is now trusted. If you later edit the config, run omni_trust again.

2. Sandbox Environment Denylist

OMNI automatically strips 50+ dangerous environment variables (BASH_ENV, NODE_OPTIONS, LD_PRELOAD, DYLD_INSERT_LIBRARIES, etc.) from all child processes. No configuration needed — this is always active.

3. Hook Integrity Verification

Custom hook scripts in ~/.omni/hooks/ are protected by SHA-256 fingerprinting.

How to use:

Add Hooks: Place your custom scripts in ~/.omni/hooks/.
Verify & Trust: Use the omni_trust_hooks MCP tool to inspect and approve the scripts. This generates SHA-256 signatures in ~/.omni/hooks.sha256.
Startup Protection: Every time OMNI starts, it re-calculates hashes and compares them to the trusted signatures.
Automatic Lockdown: If any file is modified or an untrusted file is added, OMNI will immediately exit.

Important

Modifying Hook Scripts? If you edit the content of your hook scripts, OMNI will block startup due to the fingerprint mismatch. You must run the omni_trust_hooks MCP tool again to re-authorize the updated scripts.

Best Practices: Custom Hooks

Create: Store your custom script (e.g., git-summary.sh) in ~/.omni/hooks/.
Test: Verify the script works manually in your terminal.
Authorize: Run the omni_trust_hooks tool to sign the script's current state.
Execute: Your AI Agent can now safely run the script via omni_execute:
```
{
  "command": "~/.omni/hooks/git-summary.sh"
}
```

To manually audit your hook integrity, you can run:

node dist/index.js --test-integrity

SEE SECURITY.md FOR MORE DETAILS

Performance Monitoring & Metrics

OMNI is obsessed with efficiency. Use these tools to see how much you're saving:

1. Unified Dashboard

Run the monitor to see a breakdown of tokens saved, filtering latency, and efficiency per agent:

omni monitor

Shows: Total commands processed, efficiency rating, and detailed filter/agent breakdown.

Advanced Views:

omni monitor --trend : Displays an ASCII chart of your daily distillation savings.
omni monitor --log : Shows recent tool calls and filtering results in a timeline.
omni monitor --by week : Aggregated metrics structured by week (or day, month).
omni monitor scan : Analyzes your shell history for tools that could benefit from OMNI.

2. Context Density Analysis

Measure the "Information per Token" gain for any text file or output:

omni density < build_logs.txt

Output: Calculates the exact Context Density Gain (e.g., 4.5x improvement).

The OMNI Core Pillars: Pure Intelligence

Pillar	Description	Value
Purity	Zero Semantic Loss via multi-variable confidence scoring.	Clean Signal
Density	Focus on "Information per Token" rather than simple truncation.	High Context
Speed	Zig-powered native engine with sub-millisecond response.	< 1ms Latency
Trust	SHA-256 verified project-local rules and security boundaries.	Secure
Portability	68KB universal Wasm binary runs on any runtime (Node, Web, Edge).	Universal

Market-Leading Performance

While other tools focus on simple filtering, OMNI provides a full semantic layer:

Feature	OMNI	Others
Processing Engine	Zig (Native)	Python / Go / Rust
Context Strategy	Semantic Distillation	Regex / Passthrough
Wait Overhead	Zero (<1ms)	Visible (10ms - 100ms)
Governance	SHA-256 Trust Boundary	None / Manual
Deployment	68KB Wasm / Universal	Large Native Binaries

The OMNI Advantage:

Context IQ: OMNI doesn't just shorten text; it re-writes it semantically for the LLM based on agentic intent.
Performance Supremacy: By using a persistent Wasm instance, OMNI provides instant responses without blocking the main agent execution.
Local-First Privacy: Every byte of your code and tool output stays on your machine.

Visualizing Efficiency

The "Distillation" Effect: In your AI's tool output, raw logs are transformed into a 10-line summary.
Faster Response Times: LLM processes 150x fewer tokens, giving you significantly faster replies.
Real-time Reports: Run omni monitor at any time to see the global efficiency health.
Density Metrics: Use omni density < logs.txt to calculate your exact Context Density Gain.

Installation

Homebrew (Recommended)

brew install fajarhide/tap/omni

One-Line Installer (Optimized)

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

For manual build instructions, see INSTALL.md.

Update & Uninstall

omni update       # Check for the latest version
omni uninstall    # Remove OMNI and clean up all configs

🗺️ Documentation Map

Explore the full potential of OMNI with these specialized guides:

Document	Purpose	Audience
QUICKSTART	Install and run OMNI in 60 seconds.	Everyone
CLAUDE.md	Full development guide & repo standards.	Developers
TESTS.md	Infrastructure details & how to add tests.	QA & Contributors
DSL_GUIDE.md	Create custom semantic rules without coding.	Power Users / Agents
INSTALL.md	Manual build and edge deployment instructions.	SysAdmins

Name		Name	Last commit message	Last commit date
Latest commit History 196 Commits
.agents/workflows		.agents/workflows
.github		.github
core		core
docs		docs
scripts		scripts
src		src
tests		tests
.gitignore		.gitignore
CHANGELOG.md		CHANGELOG.md
CLAUDE.md		CLAUDE.md
CONTRIBUTING.md		CONTRIBUTING.md
INSTALL.md		INSTALL.md
LICENSE		LICENSE
Makefile		Makefile
README.md		README.md
SECURITY.md		SECURITY.md
bun.lockb		bun.lockb
install.sh		install.sh
logo.png		logo.png
omni.rb		omni.rb
omni_config.json		omni_config.json
package-lock.json		package-lock.json
package.json		package.json
tsconfig.json		tsconfig.json

Uh oh!

Folders and files

Latest commit

History

Repository files navigation

The Semantic Core for Agentic AI

Why OMNI

Supported AI Agents

CLI Subcommands: Unified Intelligence

How OMNI Works

The OMNI Effect

Integration: Using OMNI Everywhere

Claude Code & Claude CLI

Codex CLI

2. Antigravity (Google)

3. OpenCode AI

Auto-Generate Config

"Always-On" Integration: Agent Autopilot

OpenCode AI Integration

Quick Setup

Supported Tools & Filters

Example Distillation

Custom OpenCode Instructions

The Adaptive Intelligence: Proxy & Distillation

1. Zero-Latency Command Proxy (--)

2. Deep Semantic Distillation (distill)

3. Ultra-Fast Benchmarking (bench)

Available MCP Tools

Easy Filtering: Zero Coding Required

2. Apply Technology Templates

Configuration Architecture

How Merging Works

Manual Configuration

Lifecycle: Creation & Editing

Security: Trust Boundary

1. Project Trust Boundary

2. Sandbox Environment Denylist

3. Hook Integrity Verification

Best Practices: Custom Hooks

Performance Monitoring & Metrics

1. Unified Dashboard

2. Context Density Analysis

The OMNI Core Pillars: Pure Intelligence

Market-Leading Performance

The OMNI Advantage:

Visualizing Efficiency

Installation

Homebrew (Recommended)

One-Line Installer (Optimized)

Update & Uninstall

🗺️ Documentation Map

License

About

Topics

Resources

License

Contributing

Security policy

Uh oh!

Stars

Watchers

Forks

Releases 18

Sponsor this project

Uh oh!

Packages 0

Uh oh!

Contributors

Uh oh!

Languages

1. Zero-Latency Command Proxy (`--`)

2. Deep Semantic Distillation (`distill`)

3. Ultra-Fast Benchmarking (`bench`)

Packages