Skip to main content

Dinoki Roadmap

Know Your Inference.
When an AI has desktop accessibility controls, it can see your screen, read windows, hear audio, and capture what’s happening. That data could be stored or analyzed without your awareness — which is why Dinoki is built around one guiding rule: privacy is not optional.

Character Mode 2.0​

Currently in development — targeting v1.4 release

Goal: Turn Dinoki’s characters into a personality-rich companion that lives on your desktop — engaging, assisting, and working alongside you, without compromising your privacy.

  • Context-aware perception – Characters can read screens, track windows, and hear audio, but only share the minimum data required for the task.
  • Transparency – You’ll always know exactly what’s sent to inference providers.
  • Goal tracking – Characters can create and track their own goals, add new ones based on conversation and history, and let you define “usefulness” in the system prompt.
  • Interactive feedback – Quick multiple-choice prompts to unblock tasks or keep you engaged.
  • Autonomous tool use – Characters can run a small, controlled set of tools independently in parallel with you.

Character Mode 2.0 is designed to be fun, responsive, and helpful — a desktop presence that feels alive but never invasive.

Local-First Architecture (Fully Offline Capable)​

Today, most local models are either not powerful enough or require more compute than typical desktops provide. As models improve and hardware gets faster, we expect offline adoption to grow significantly.

Goal: Make Dinoki as powerful offline as it is online.

  • Download and run AI models directly inside Dinoki.
  • Support Apple’s upcoming on-device LLM.
  • Improve performance for smaller local models like Qwen3.

Tools & Ecosystem Growth​

The MCP ecosystem is growing but still early. Better standards and patterns will make adoption easier. As it matures, Dinoki will act as both an MCP Client and MCP Server, allowing other AI tools — including cloud-based ones — to run in local-first environments for greater security and control.

Goal: Give Dinoki access to a wide, ever-expanding toolbox.

  • Add more built-in Dinoki tools.
  • Integrate with open MCP tool directories.
  • Open-source selected tools for community contributions.

Agent Mode 2.0​

Long-term vision — will launch only after privacy-first architecture is complete.

Goal: Build a complete desktop automation suite — a digital copy of you — capable of operating your Mac and working towards your goals, hands-free.

  • Full desktop control – Read inputs (keyboard, mouse, screen, audio) and generate outputs (clicks, keystrokes, file edits, app actions).
  • Goal-oriented execution – Break down your objectives into actionable steps and carry them out without constant supervision.
  • Context-aware decision making – Adapt actions based on the current environment, app states, and evolving priorities.
  • Complex workflows – Automate multi-step tasks, from data gathering to document creation to app management.
  • Privacy-first foundation – All automation runs within strict data boundaries so your personal information stays private.

Agent Mode 2.0 is designed to feel like a true digital version of you — able to perform your work, follow your priorities, and adapt to changes — while always keeping you in control.

We’re building Dinoki for the long run.
More autonomy, more capability — always with privacy first. Dinoki belongs to you, not the cloud.