Less noise.
More signal.
OMNI is a smart terminal layer that intelligently filters and prioritizes command output before it reaches your AI agent.
Cut token consumption by up to 90% and give your AI the clean context it needs to reason better.
Fully transparent. You're always in control.
brew install fajarhide/tap/omni The Two Pillars of OMNI
Making AI agents work better for longer, without wasting tokens.
Pillar 1: Token Efficiency
Every byte that enters the context window must be meaningful. OMNI acts as a smart sieve, ensuring zero tokens are wasted on useless output.
- Build Outputs: Compresses a 3,000 token compile error into just 80 tokens of pure signal.
- Test Runners: Filters out 4,000 tokens of passing tests to show only the exact assertion failures.
- DevOps & Infra: Transforms thousands of lines of Terraform/Kubectl output into a concise status summary.
Pillar 2: Context Fidelity
High compression is useless if the AI hallucinates. OMNI ensures your agent never loses its train of thought, even after 4 hours of coding.
- Context Pressure Management: Proactively warns the agent before the context window crashes.
- Critical File Pinning: Permanently pins
AGENTS.mdor.cursorrulesso core instructions are never forgotten. - Graph Indexer: Analyzes dependency graphs to warn agents when they mutate high-impact, frequently imported files.
Feel the Difference
See exactly how much semantic space you save.
Real-World Use Cases
How OMNI solves the daily frustrations of Agentic AI developers.
The "Infinite Loop of Death"
Scenario: AI gets distracted by 20,000 lines of npm install warnings, missing the actual build error at the end.
OMNI's Fix: Mutes all peer dependency warnings and surfaces only the exact Build Error with its stack trace. AI fixes the code instantly.
The "Silent Hallucination"
Scenario: AI reads a 3,000-line file via cat, struggles to keep it in working memory, and starts hallucinating.
OMNI's Fix: Replaces raw cat with a Structured Outline. Shows only imports, public APIs, and risk markers—reducing output by 80%.
Multi-Agent Collaboration
Scenario: Cursor IDE and Claude CLI run simultaneously, executing redundant commands and wasting tokens.
OMNI's Fix: Acts as a shared memory layer. Cursor and Claude share the same filtered memory streams, active errors, and environment.
The Breakthrough Layer
A native, invisible utility belt explicitly built for the Agentic AI era.
Distill Monitor
Run omni stats to track lifetime token savings and ROI across all your projects.
Session Intelligence
OMNI tracks hot files and active errors across your workspace. It remembers what the AI knows, injecting contextual continuity into every invocation to save even more tokens.
RewindStore - Zero Signal Loss
Worried OMNI filtered something important? Built-in RewindStore saves raw output locally via SHA-256. If the AI needs the full log, it can retrieve it instantly.
Omni Learn - Autonomous Discovery
Analyze your terminal history to identify noise patterns and generate optimized semantic filters automatically.
Omni Doctor: Self-Healing Config
A proactive health check utility that scans your filters, dependencies, and environment for issues, providing instant --fix solutions.
Universal Support (15+ Agents)
Built with a Native Rust Core for sub-millisecond execution. OMNI seamlessly integrates with Claude Code, Cursor, Cline, Pi, Copilot, Antigravity, and exposes 12 MCP Tools.
Command-First Architecture
The world's first context-aware terminal signal pipeline. Built for the Agentic era.