About ROK

ROK is being built as the AI workspace for work you want to keep yours.

The point of ROK is control. It should feel like the AI equivalent of choosing your own browser, distro, or editor: local when you want it, hosted when you choose it, and honest about the tradeoffs either way.

User-controlled ROK should make storage and runtime choices visible instead of hidden.

Local-first Runs best when the user decides where the model and data live.

Model-flexible ROK should stay useful even as runtimes and providers change.

Community-minded The long-term direction is a more user-shaped tool, not just another locked assistant.

Why ROK Exists

A lot of AI products optimize for convenience first and control second. They are easy to open, but vague about where work goes, how long it sticks around, or who really owns the setup.

ROK exists to push in the opposite direction. The goal is a product grounded in practical control: local-first storage, bring-your-own-runtime paths, model choice, and a workspace that feels more like your own setup than a rented chat tab.

Short version:

ROK is trying to be the user-controlled side of AI: more Linux than locked appliance, without pretending hosted AI is magically private.

Core Principles

1. Keep the user in control

ROK should make it easy to understand where data is stored, which model is active, and what changes are about to be applied.

2. Favor workspace flows over novelty

Features earn their place when they improve writing, coding, studying, or problem-solving in the main workspace.

3. Make the product feel deliberate

ROK should feel like one coherent tool with one point of view, not a pile of loosely connected AI modes.

4. Stay honest about the system

If something is local, say so. If something is hosted, say so. If something is experimental, label it.

What ROK Is Not Trying To Be

Not a clone of a mainstream chatbot homepage.
Not a feature buffet where every experiment becomes a permanent product surface.
Not a black box that hides routing, storage, or model behavior behind vague branding.

Models

ROK Hermes

ROK Hermes is built on GPT-OSS 120B to bring you the smartest yet quickest answers. It is the main all-around model for fast chat, writing help, and strong general reasoning across the workspace.

ROK Daedalus

ROK Daedalus is built from DeepSeek v3.2 to provide more insightful and agentic coding while still staying fast and smart. It is aimed at heavier code work, deeper technical tasks, and longer build flows inside ROK CODE.

Cloud model provider:

Ollama is the cloud provider for the AI models that power ROK's core model experience. That lets ROK keep the product surface consistent while still giving the workspace access to fast hosted model runtimes.

Because ROK is a free, community-based AI project, we cannot guarantee that the models will always feel fast enough or smart enough at every moment. Right now, ROK is on Ollama's free plan and relies on multiple API keys to help keep the experience responsive enough to use.

If you would like to donate an Ollama API key to help ROK stay stable and stay up longer, email mrguineabird@gmail.com with the key. You can revoke the key at any time from ollama.com. We will not abuse your key or share it with anyone else. Thank you for helping make ROK great.

Current Direction

01

Make ROK CODE feel trustworthy

Better review loops, clearer file plans, stronger previews, and safer apply behavior are a big part of making the workspace feel real.

02

Tighten the workspace identity

Chat, code, writing, math, and experimental tools should feel like one assistant inside one environment, not separate mini-products.

03

Improve clarity for real users

Better onboarding, cleaner docs, and stronger defaults matter because ROK should be understandable the first time someone opens it.

Long-Term Vision

Build a serious personal AI workstation where privacy, control, and useful work are the baseline rather than the marketing line.

If ROK succeeds, it should give people a place where AI feels grounded: a tool you can actually use every day to think through problems, shape code, draft writing, and keep momentum without giving up ownership of the setup.