zeraix/zeraix

What it solves

Zeraix provides a unified desktop workspace for interacting with AI models, files, and tools. It solves the fragmentation of local AI by integrating model management, file system access, and agentic task execution into a single application, allowing users to run powerful models on their own hardware without needing constant cloud connectivity or subscriptions.

How it works

Zeraix acts as an orchestrator that combines a local inference core (using llama.cpp-based runtimes) with a Rust-based Agent Runtime. This allows the AI to read, search, and edit project files, execute terminal commands, and use browser tools. It supports GGUF models and OpenAI-compatible endpoints, and offers an optional QEMU sandbox to isolate command execution for better security. Users can control the AI's autonomy through different approval modes (Default, Full trust, Manual approval, or Plan mode).

Who it’s for

It is designed for developers and power users who want a local-first AI environment to manage ongoing projects, automate multi-step workflows, and maintain privacy by keeping conversations and inference on their own device.

Highlights

• Local-First Core: Run GGUF models locally with hardware-aware recommendations and no required account.
• Agentic Capabilities: Delegate tasks to specialized sub-agents with granular control over tool approval and execution.
• Integrated Tooling: Built-in support for Word, PDF, spreadsheets, and presentations, plus integration with MCP servers and plugins.
• System Integration: Direct access to project files, terminal commands, and a media library for generated assets.
• Security Layers: Optional QEMU sandbox for isolated command execution and multiple permission levels for protected actions.