Changelog

System Prompts

Onit now allows you to customize your system prompt! Hover over the carrot next to the "new" button in the top toolbar, and you can set up a new System Prompt to match your workflows.

Streamed Responses

v1.2 Introduces one of our most requested features: streamed responses! Like other AI tools, Onit now streams responses word-by-word instead of waiting for the whole response to generate. This creates a much smoother, quicker experience!

Other Misc

• New Loading indicator

• Option to disable the 'ESC' shortcut (PR #106)

• Clicking the top-right 'hint' view opens the Onit panel (PR #91)

• *Experimental* Setting to start a new chat each time the dialog is dismissed (PR #82)

• *Experimental* Setting to use launch shortcut as a toggle (PR #81)

• Add the Model used to the reponse view (PR #77)

• Enable copying from send prompts (PR #61)

• Option to adjust line height (PR #74)

• New App Icon

Plus, lots of great bug fixes!

Big thanks to @mafikes and @Abhinandan-Khurana for their feedback on this release!

✨ New Features

Auto-Context
Onit can now automatically pull relevant text from your computer when launched, using Apple's Accessibility APIs:

• Auto-Context from Highlights: If text is highlighted before launching Onit, it’s added as context to your prompt.

• Auto-Context from Current Window: Launching Onit with a new shortcut (Cmd+Shift+0 by default) pulls text from the active window into your prompt.

——

⚠️ Privacy Note: Auto-Context data is not uploaded unless you submit a chat. In remote mode, context is sent directly to the model provider, not third-party intermediaries. In local mode, nothing is uploaded.

💎 Improvements

New Model & API Support

• Deepseek: Added support for Deepseek as a model provider.

• Custom Local Endpoints: Specify a custom HOST and PORT for Ollama when running it on a separate machine.

• Custom Remote Providers: Connect to any remote provider that follows the OpenAI API schema by specifying a custom URL and TOKEN (e.g., OpenRouter, Groq, Hyperbolic).

Local Mode Enhancements
Added advanced parameters for local inference, including: timeout, keep_alive, context_window, temperature, top_k, top_p.

Customization Options

• Window Size & Positioning: Onit now remembers the last-used window width and lets you choose between top-right, top-center, or top-left placement.

• Custom Font Size: Adjust font size in settings.

💡 Bug Reports & Feedback
Accessibility APIs can be finicky! If you notice any unexpected behavior, please open an issue on GitHub.

We're thrilled to announce the launch of Onit, a ChatGPT Desktop application designed to make your AI interactions more accessible and enjoyable!

✨ Features

• Local Mode: Engage with AI models directly on your computer without needing an internet connection.

• Multi-Provider Support: Choose from top AI models, including OpenAI, Anthropic, xAI, and GoogleAI, to find the perfect fit for your needs.

• File Upload: Easily add images or files to your conversations with simple drag-and-drop functionality.

• History: Revisit past chats effortlessly using the history view or the convenient up/down arrow shortcut.

• Customizable Shortcut: Set your preferred hotkey to launch the chat window, with Command+Zero as the default.

Onit is open-source and available for download on macOS. You can also access the source code on GitHub. It's licensed under a Creative Commons Non-Commercial license, allowing you to customize Onit to your needs while ensuring it's used responsibly.

🔮 Future plans

We're excited to share our upcoming plans for Onit, aiming to enhance your AI experience even further:

• Auto-context: Automatically retrieve context from your computer, eliminating the need for repeated uploads.

• Local-RAG: Empower users to index and create context from their files without uploading anything.

• Local-typeahead: Implement a feature similar to Cursor Tab, but accessible everywhere.

• Additional Support: Expand compatibility to include Linux and Windows, as well as models like Mistral and Deepseek.

• Bundle Ollama: Integrate Ollama to streamline the download process.

And that's just the beginning! We have many more exciting features in the pipeline to make Onit even more powerful and user-friendly

We hope Onit enhances your AI experience and look forward to hearing your thoughts!