Run specialized 1B/3B brains (custom-trained models) entirely on your machine. No API keys. No cloud. No monthly fees.
curl -fsSL https://roe-defense.github.io/omni/install.sh | bash
⚡ Optimized for Apple Silicon (Metal) & NVIDIA (CUDA)
Omni is a Secure AI Stack for developers who value privacy, speed, and precision.
Cloud LLMs (like GPT-4) are powerful but unsafe. They leak your IP, suffer from latency, and cost money per token.
Omni takes a different approach: Hyper-Specialization. Instead of one giant model, we give you a swarm of small, expert "Brains" (1B/3B parameters) that run locally on your hardware. They are faster, smarter at their specific niche, and entirely free to run.
Designs scalable systems and infrastructure.
Builds robust APIs, Databases, and Logic.
Crafts responsive UIs and User Experiences.
Don't just use our intelligence. Clone your own.
Feed Omni your private documents, codebases, and notes. It trains a custom LoRA adapter entirely on your machine.
Ingest PDF, Markdown, TXT, and source code. Perfect for legacy code refactoring, legal document analysis, or personalized writing assistants.
Soon, Omni will see and hear. Train on video walkthroughs, architecture diagrams, and voice meeting logs. The ultimate context window is reality itself.
omni train --name @my/secret-project --path ./docs
Select a specialized brain to load into your stack.
Expert in system design, cloud infrastructure patterns, and technical strategy. Starts your project right.
Writes robust Python (FastAPI), Node.js, and SQL. optimized for async patterns and database schemas.
Builds responsive UI components in React, Vue, and pure HTML/CSS. Masters Tailwind utility classes.
Automates deployment. Generates Dockerfiles, K8s manifests, and CI/CD pipelines.
Audits code for vulnerabilities, configures firewalls, and hardens servers.
The Command Line Assassin. Translates English to complex bash/zsh/awk one-liners.
The Query Optimizer. Writes raw SQL queries for PostgreSQL and ANSI SQL.
The Version Controller. Handles rebase, bisect, and complex merge conflicts.
Native iOS engineering. Swift, SwiftUI, and Core Frameworks.
Native Android engineering. Kotlin, Jetpack Compose, and NDK.
Cross-platform UI with Dart. Riverpod state management and native bridging.
TypeScript-based native apps. Expert in Reanimated, Expo, and Native Modules.
C# scripting and Unity Editor logic. Prefabs, Physics, and DOTS.
C++ actor logic and Blueprint graphs. Engine API mastery.
Android Team Awareness Kit plugin dev. Map overlays, custom tools, and sensor integration.
Windows TAK plugin dev. C#/.NET based TOC tools and heavy data visualization.
iOS Team Awareness Kit plugin dev. Swift/SwiftUI for mobile-first situational awareness.
Infrastructure admin. SSL cert management, Federation config, and Data Packages.
Operator Q&A. "How to pair drone?", "Broadcast 9-line", and field troubleshooting.
Cursor on Target (CoT) Protocol generator. Strict XML schemas for sensor feeds.
Native macOS apps with Swift and AppKit. System extensions and XPC.
Modern Windows development. C#, .NET 8, and WinUI 3.
Designs and optimizes LLM pipelines, RAG systems, and dataset generation scripts.
Use @roe/architect to design the schema, @roe/backend to write the API, and @roe/frontend to build the landing page. A full MVP in hours, not weeks.
Working on sensitive IP or classified code? Omni runs without an internet connection. Your code never touches a cloud server.
Train a custom brain on your legacy codebase documentation. Then ask it to refactor old functions into modern patterns.
Spin up a fleet of specialized agents. Have @roe/ai-eng design the prompt strategy while @roe/backend builds the execution harness. Self-improving loops running entirely on local hardware.
Chat with your sensitive legal documents, medical records, or proprietary research. No API fees, no data leaks. Just point Omni at your PDF folder.
Single brains are powerful. Connected brains are unstoppable.
The Swarm Bus allows specialized agents to collaborate on complex tasks without human intervention.
@roe/architect designs the system, then passes the spec directly to @roe/backend and @roe/frontend to execute in parallel.
A unified vector store allows all agents to access the same project context, documentation, and decision history.
If @roe/backend's API doesn't match @roe/frontend's needs, they negotiate and fix the interface automatically.
Mobile, Desktop, and AI Engineering brains. "Train Your Own Brain" (Local LoRA) wizard.
Real Inference (Llama-3B), Code Execution, and the "Secure" TUI. Live now.
Multi-agent bus. Brains talk to each other to solve complex tasks. Shared Vector Memory.
Screen Awareness (Vision) + Voice Mode + Air Gap Kill Switch.
Defense Suite (TAK), Desktop (Mac/Win), and Game Engines (Unity/Unreal).
Certified Zero-Telemetry. Official Hardware Kits. The Hive P2P Mesh.