zeraix/zeraix

Zeraix – A desktop‑first AI workspace

What it is – Zeraix is an open‑source desktop application that lets you run large language models (LLMs) and AI agents entirely on your own computer. It bundles model download/management, a chat‑style assistant, a developer‑mode that can read/write files and run terminal commands, and a scheduler for recurring AI workflows. The core runtime currently uses the community‑maintained llama.cpp engine, while the team’s own research runtime ExactFlux is being developed for future releases.

Why it matters – Most AI products are built around cloud APIs; Zeraix flips that model, making the local model the default. This gives you:

• Privacy – prompts, files and conversation history never leave the device unless you explicitly use a cloud endpoint.
• Offline capability – after the model and runtime are installed, everything works without an internet connection.
• Hardware‑aware optimisation – the app inspects your CPU/GPU/Unified Memory and recommends quantized GGUF models that fit, plus it applies research‑grade optimisations (mapped weights, speculative decoding, KV‑cache reuse, etc.) for faster, lower‑memory inference.

Key components

Component	What it does
Zeraix Desktop	GUI for installing GGUF models, launching a local OpenAI‑compatible endpoint, chatting, and managing scheduled workflows.
Assistant Mode	Conversational UI that can handle text, images and multimodal models; supports “Skills” (pre‑built tool extensions) and optional cloud models.
Developer Mode	Gives the selected model controlled access to a user‑chosen workspace: file search/edit, diff preview, terminal execution, browser tools, and delegation to sub‑agents.
Scheduled automation	Define recurring AI jobs (daily, hourly, etc.) with run‑history, notifications and per‑run output folders.
QEMU sandbox	Optional lightweight virtual machine that isolates agent commands, with per‑command filesystem scoping via bubblewrap.
Model Systems research	Ongoing internal work (ExactFlux) on memory‑efficient inference, speculative decoding, MoE handling, and hardware‑specific back‑ends (Metal, CUDA, Vulkan).

Supported platforms – macOS 13+ (Apple Silicon) and Windows 10/11 x64. The public release ships a pre‑built llama.cpp runtime for each platform; ExactFlux is not yet included.

Typical workflow

1. Download the installer from the GitHub releases page.
2. Install the app (drag‑and‑drop on macOS, standard .exe installer on Windows).
3. Add a model – the Model Library scans your hardware, suggests a compatible GGUF model, and downloads it together with the matching llama.cpp binary.
4. Start a conversation – select the local model in the Assistant and type a prompt; inference runs locally.
5. (Optional) Enable Developer Mode – point the app at a project folder so the model can read, edit, and execute code under your explicit approvals.
6. (Optional) Create a scheduled workflow – define a prompt or script that runs automatically on a chosen schedule.

Who should consider Zeraix

• Developers or researchers who want a self‑contained AI environment without relying on paid cloud APIs.
• Teams that need data privacy (e.g., working with proprietary code or confidential documents).
• Users with a decent amount of RAM (≥ 16 GB recommended) and a modern CPU/GPU who want to experiment with quantized LLMs locally.
• Anyone interested in the cutting‑edge research on making large models run efficiently on consumer hardware (the team publishes performance measurements and research updates).

Maturity & licensing

• The desktop application is at v1.11.0 (stable public release). Core features—model management, assistant, developer mode, scheduling, and sandboxing—are production‑ready.
• The research runtime ExactFlux is still “Exploring/Prototype” and not part of the public source tree; it will be open‑sourced once validated.
• Licensed under Apache‑2.0, so you can use, modify and redistribute the code under the same terms.

Getting started

• macOS: download the .dmg, drag to Applications, open (allow the security warning if shown).
• Windows: download the .exe, run the installer, launch Zeraix.
• Follow the built‑in “Quick Start” wizard to install a recommended GGUF model and begin chatting.

Community & support

• Discord: https://discord.gg/PcQ3jr3MfH
• X (Twitter): https://x.com/ZeraixAI
• Issues & feature requests are handled via the GitHub Issues page.

---

Bottom line – Zeraix provides a polished, privacy‑first desktop environment for running local LLMs, with a strong emphasis on system‑level performance research. If you want to experiment with powerful models on your own hardware without a cloud subscription, Zeraix is a concrete, production‑ready option.