Choose OpenRouter Free for page chat and summaries without buying OpenRouter credits. Use your existing connection or sign in from the sidebar. Daily request limits apply. OpenRouter Auto remains available for paid models, web search, and Agent Mode. Your AlliHat subscription and trial are unchanged.

A redesigned welcome screen explains the trial up front, then walks you through enabling AlliHat in Safari and choosing your AI. Pricing and the 7-day free trial are unchanged.

AlliHat now works directly with Grok and Perplexity. Bring your own xAI or Perplexity API key for page-aware chat, web search, PDFs, and Agent Mode. This release also keeps slow Grok requests alive more reliably when Safari suspends extension background work.

AlliHat now works with OpenRouter. Connect your OpenRouter account in one click, or paste an existing key, and let OpenRouter choose the model for each task. OpenRouter supports chat, web search, and Agent Mode.

ChatGPT now searches the web before saying it does not know when an answer depends on missing or current information. It also spends more time reasoning through each request.

Agent Mode now works with ChatGPT and Gemini. Pick the model you want, then let AlliHat click buttons, fill forms, scroll, and navigate. Also improved: when OpenAI or Gemini rate limits a task, AlliHat now shows the provider's actual error instead of a generic retry message.

Search your chat history. Open Chat History in the sidebar, or the AlliHat home page, and start typing to filter your past conversations. It searches inside the conversations, not just the titles, so you can find a chat by a phrase you remember from the middle of it. Matches are highlighted, and when the match is buried in a long chat you get a one line excerpt showing where it hit. Press / on the home page to jump straight to the search box. Also fixed: long titles in the sidebar history list could lose their trailing ellipsis, and the list could scroll sideways by a hair.

Agent mode now handles longer tasks (like a full shopping list) without stopping early. Replaced the fixed step limit with a smarter budget that keeps going as long as the agent is making real progress, and stops quickly if it gets stuck.

Agent mode improvements: - Fill without submitting. Ask AlliHat to fill out a form and it stops there so you can review it. It only submits when you tell it to. - More reliable form filling on modern, interactive websites. - Better at recognizing the buttons, links, and fields on a page. - Fixed an error that could cut off longer agent tasks. - Returning to the page you started on no longer shows an extra confirmation.

AlliHat now reads PDFs. Open a PDF in Safari and ask for a summary or dig into specifics. The whole document goes to your model (Claude, ChatGPT, or Gemini), so it handles tables, layout, and scanned pages, not just plain text. YouTube summaries are also more reliable. When the transcript panel is slow to load, AlliHat waits it out and retries instead of reporting that it couldn't find one.

AlliHat 3.0 adds your choice of AI. Alongside Claude, you can now use ChatGPT or Gemini with your own API key and switch between them anytime. On Macs with Apple Intelligence, you can start free with on-device AI, no key required. This release also includes a faster default model setup and reliability fixes when switching models.

Minor under-the-hood improvements and stability tweaks.

Added a usage panel in the Model picker. See your estimated Claude API spend for this month and all time.

- New right-click actions on selected text: Explain, Summarize, Synonyms, or Ask about this. - Select text, then click the AlliHat icon, and your selection appears in the sidebar.

AlliHat now supports Apple's on-device model. On macOS 26+ Macs with Apple Intelligence enabled, new users can pick "Apple Intelligence" instead of Claude on first launch — completely free, no API key, runs entirely on your device. Existing users with their Claude API key already set up are unaffected. Switch between Claude and Apple Intelligence anytime from the Model menu.

Updated system requirements and improved compatibility with older macOS versions (now supports macOS 12+).

Bug fixes