All releases

Apps, your plan, and a voice

Apps you place where you want them, a ChatGPT plan in place of a key, model servers on your Mac, dictation, and one screen for every permission.

This release gives Apps a place of their own, adds two more ways to bring a model, and lets you talk to Alice.

Apps

An App is one file, documents/apps/<name>.html. It opens in its message, beside the conversation, full screen, or hidden while it keeps running. ‘Apps’, at the top of the window, lists every App and the ones that are running. Alice checks an App before she saves it, can read its state, and can press its actions for you.

More ways to bring a model

Sign in to OpenAI with a ChatGPT Plus or Pro plan and chat runs on that plan; voice, web search and images on OpenAI still need an API key. A model server on your Mac or your own machine, such as LM Studio, connects by its address. A subscription comes with Included AI, which needs no key.

Voice

Dictate a message from the prompt and have a reply read aloud, with a voice for each agent. Both use an OpenAI or Gemini key.

Services

Connect a remote MCP server by its address, or a web service from its API documentation. Each call shows a card first, and each MCP tool has its own switch.

Permissions

Settings has a ‘Permissions’ screen: every ‘Always’ you gave, grouped by what it reaches, with ‘Remove’ beside it.

Try it

Ways to connect a model covers the plan sign-in and model servers, and What her code and her apps can do covers Apps.