Single Native Rust Binary • Zero Node/Python Dependencies

The High-Performance, Path-Jailed AI Agent Harness

`ox` is an uncompromising coding agent harness built in pure Rust. Featuring kernel-level PathJail sandboxing, automatic subprocess environment secret scrubbing, non-destructive DAG session branching, and universal LLM provider streaming.

curl -fsSL https://raw.githubusercontent.com/JohnnytheShark/ox-orchestrator/main/install.sh | bash
~12 MB
Single Static Binary
< 15 ms
Cold Start Time
Zero
Runtime Dependencies
100%
Path-Jailed Safety
ox chat — session-a1b2c3d4
LIVE SIMULATOR
DAG Session Topology:
─── ─── ┬──

Engineered For Autonomous Safety & Speed

Traditional AI agent harnesses rely on bloated node runtimes, leaky child processes, and fragile linear histories. `ox` redefines the agent harness from first principles in Rust.

Native Rust Performance

Compiles to a standalone ~12MB native binary with LTO, abort panic, and zero external runtime dependencies. Instant cold starts under 15ms.

Kernel-Level PathJail

Every file access is strictly jailed to the authorized project root via rigorous canonical path checking, preventing symlink traversal and climbing attacks.

Subprocess EnvScrubber

Child subprocesses and MCP tools inherit zero sensitive environment variables. API tokens (*_API_KEY, *TOKEN*, *SECRET*) are scrubbed before fork.

Non-Destructive Session DAG

Never lose prior context. Conversational turns form an immutable directed acyclic graph. Branch, rewind (/undo), and checkout alternative reasoning trails safely.

Model Context Protocol (MCP)

Native JSON-RPC 2.0 stdio client integrates external tools effortlessly (SQLite, GitHub, Puppeteer, file systems) with sandboxed tool execution.

Zero-Memory Secret Zeroization

Sensitive credentials in memory implement Rust's zeroize::Zeroize trait, clearing sensitive key bytes with compiler memory fences when dropped.

Structured by the Diátaxis Framework

Clear, purposeful technical documentation organized into four user-centered quadrants: learning, problem solving, information lookup, and deep architectural rationale.

LEARNING-ORIENTED Practical Acquisition

Tutorials

Step-by-step guided lessons for newcomers to take their first successful steps with `ox`.

PROBLEM-ORIENTED Practical Application

How-To Guides

Task-focused recipes to solve real-world problems in configuration, security, and CI/CD.

INFORMATION-ORIENTED Theoretical Application

Reference

Authoritative, comprehensive technical facts, CLI specs, configuration schemas, and security contracts.

UNDERSTANDING-ORIENTED Theoretical Acquisition

Explanation

Deep background discussion, architectural philosophy, design trade-offs, and comparison studies.

Diátaxis Explorer 14 Guides
Tutorials / Getting Started

Zero Vendor Lock-In

Seamless streaming SSE integrations for frontier cloud LLMs and local offline inference engines.

Recommended

Anthropic

Claude 3.7 Sonnet, Claude 3.5 Sonnet, Haiku

export ANTHROPIC_API_KEY="..."
ox chat -m claude-3-7-sonnet-20250219
First-Class

OpenAI

GPT-4o, GPT-4o-mini, o1, o3-mini

export OPENAI_API_KEY="..."
ox chat --provider openai -m gpt-4o
Supported

Google Gemini

Gemini 2.0 Flash, Gemini 1.5 Pro

export GEMINI_API_KEY="..."
ox chat --provider gemini
Supported

DeepSeek

DeepSeek-V3, DeepSeek-R1 (OpenAI compat)

export DEEPSEEK_API_KEY="..."
ox chat --provider openai -m deepseek-chat
Offline & Local

Ollama

Qwen 2.5 Coder, Llama 3.3, DeepSeek R1 local

# Zero API key required
ox chat --provider ollama -m qwen2.5-coder:14b

Download Precompiled Binaries

Standalone, optimized machine binaries with Link-Time Optimization (LTO) and zero external runtime dependencies.

v0.1.0 Latest Release View All Releases on GitHub ↗

Linux x86_64 / ARM64

glibc & musl static

macOS Apple Silicon

Native ARM64 / M-Series

Supports macOS 12+ (Monterey, Ventura, Sonoma, Sequoia).

Windows x64

Standalone .exe

Supports Windows 10, Windows 11, and Windows Server.

Build via Cargo

Source Distribution

Compile the latest version directly with your local Rust toolchain.

cargo install --git https://github.com/JohnnytheShark/ox-orchestrator ox-cli
Command copied to clipboard