Thuki
local Al overlay for macOS (double-tap Control, no API key)
Last verified:
What is Thuki?
Thuki is a floating AI assistant for macOS that appears directly over any app with a double‑tap on the Control key, designed to keep you in your flow without tab‑switching. It uses local AI models (via Ollama) running on your own machine, so prompts and answers are processed locally and do not require an account, email, or card. The interface is minimal: highlight some text, take a screenshot, or type a slash command, get an answer, then dismiss the overlay and continue working.
Key features include a global hotkey that works even in fullscreen apps, automatic pre‑filling of selected text as context, and built‑in commands such as /screen to capture your desktop, /search for agentic web lookups, and /think for step‑by‑step reasoning. Thuki also supports slash‑style verbs like /translate, /tldr, /rewrite, /bullets, /todos, and /explain, letting you reshape or summarize content without leaving your current window. The entire conversation history is stored in a single SQLite file on your disk, with no external server or cloud logging.
Thuki targets macOS users who want a lightweight, private, and fast AI secretary for quick, throwaway conversations: developers, designers, writers, and power users who already run local models with Ollama. It is especially useful when you frequently need instant help with code, documentation, or on‑screen visuals without context‑switching into a separate AI app. The tool is tuned for Apple Silicon Macs where the local model can run efficiently on the GPU, making it ideal for anyone who values privacy, offline‑first workflows, and tight integration with their existing desktop environment.
Long‑term, Thuki is being shaped toward a more agentic assistant that can perform tasks across tools like Gmail, Calendar, Slack, and others via MCP integrations, going beyond just answering questions to actually doing small workflows on your behalf. However, the current focus remains on fast, local, context‑aware interactions that respect your attention and data. The project is open source under the Apache 2.0 license, and the app is offered at no cost, with no subscription or recurring billing.
Thuki is for macOS‑only, currently built and tested on macOS 13 and later, and it relies on you installing and managing a local model runtime such as Ollama separately. This means your data stays on your machine, your model runs on your hardware, and you can sandbox the model further using optional Docker‑based isolation if desired. The combination of a floating overlay, simple slash commands, and fully local inference makes Thuki a minimalist but powerful companion for everyday desktop work without the friction of cloud‑based AI tools.
Thuki pricing
Pricing model: Freemium
Thuki is free to use with no subscription, no recurring billing, and no account or payment required. After downloading the app you run it locally on your Mac, and the model runs on your own hardware via a local runtime like Ollama, so there is no per‑token or per‑query charge. The only cost is the hardware and time it takes to download and run your chosen local model; Thuki itself does not add any additional fees.
Thuki pros
- Runs entirely on your own machine with local models
- No account, email, or payment required
- Floating overlay appears above any app with double‑tap Control
- Works even in fullscreen apps
- Pre‑fills selected text as context automatically
- Built‑in /screen command to capture your full desktop
- Supports /search for agentic local web search pipeline
- Offers /think for step‑by‑step reasoning chains
- Slash‑style verbs for quick tasks like /translate, /tldr, /rewrite
- Minimal interface with no menus or complex settings
- Open source under Apache 2.0 license
- Conversations stored in a single SQLite file you control
- No cloud logging or external server tracking
- Offline‑first; works without internet after model is downloaded
- Compatible with popular open models such as Llama 3, Mistral, Phi‑3, Qwen, and Gemma
- Can optionally run the model in a sandboxed Docker container
- Tuned for Apple Silicon Macs and their GPU capabilities
- Does not bill you for usage once installed
- No per‑token or per‑query pricing
- Maintains a small, focused workflow for quick throwaway questions
Thuki cons
- macOS‑only, not available on Windows or Linux
- Requires separate installation and setup of a local model runtime like Ollama
- Limited to devices that can run local models efficiently (especially older or non‑Apple Silicon Macs)
- No built‑in cloud‑hosted model switching; you must manage models yourself
- No native mobile app or cross‑device sync
- Relies on macOS permissions (Accessibility and Screen Recording) which some users may be cautious about
- No full‑featured task automation yet across apps like Gmail or Slack
- No built‑in multi‑model orchestration or easy model‑switching UI
- No advanced UI preferences or appearance customization
- Does not store or index your entire disk or project by default, limiting deep context across files
Frequently asked questions about Thuki
What is Thuki and what does it do?
Thuki is a floating AI assistant for macOS that appears over any app when you double‑tap the Control key. It sends your questions and context to a local AI model running on your machine and returns answers directly in the overlay, aiming to minimize context‑switching and keep you focused on your current work.
Does Thuki require an account or subscription?
No, Thuki does not require an account, email, subscription, or payment card. You install the app once, connect it to your local model runtime such as Ollama, and use it freely without any recurring billing or cloud service fees.
Where does my data live when using Thuki?
Your conversations and data live in a single SQLite file on your local disk, with no external backend or server. That file is under your control; deleting it removes the stored conversations, and there are no hidden logs or cloud‑stored copies of your prompts.
Can Thuki work offline?
Yes, after you download and set up a local model with your chosen runtime such as Ollama, Thuki can operate entirely offline for everyday prompts because the model runs on your own hardware. Web‑related features like /search will need internet access when you use them.
Which AI models can Thuki use?
Thuki uses the local models you run via Ollama or similar local runtimes, including open models such as Llama 3, Mistral, Phi‑3, Qwen, and Gemma. You pick which model to load and configure it in your model runtime; Thuki simply sends prompts to that local instance and displays the responses.
How does Thuki integrate with my current workflow?
Thuki integrates by letting you double‑tap Control from any app, including fullscreen windows, to summon the floating assistant. Selected text is automatically used as context, and you can type questions or slash commands like /screen, /search, or /think without leaving your current document, terminal, or browser.
Is Thuki open source?
Yes, Thuki is open source under the Apache 2.0 license. The source code is available on GitHub, and the project is built as a transparent, community‑driven tool rather than a closed cloud service.
Does Thuki ever send my data to the cloud?
By default, Thuki does not send your data to any external cloud service; inference runs on your own machine via your local model runtime. Only explicitly enabled web tools such as /search reach the internet when you invoke them, and that is handled locally through your own pipelines.
Which macOS versions does Thuki support?
Thuki is built for macOS 13 and later, and it is optimized for Apple Silicon Macs where the local model can run efficiently on the built‑in GPU. You should check the GitHub releases page for the current supported builds and compatibility notes.
Can I sandbox the AI model when using Thuki?
Yes, Thuki supports optional sandboxing of the model runtime, for example via a hardened Docker container that drops unnecessary capabilities, mounts weights as read‑only, and restricts network access to localhost‑only. This feature is opt‑in and not enabled by default, giving you a choice to run the model in a more secure environment if desired.