Choose and open your AI
Talk to AI uses the choice saved in OmniAgents.
Press “Talk to AI”. With no active choice, the window shows Claude, ChatGPT and Local model (private).
Press “Connect” below the desired tile. After choosing, only the selected AI’s tile remains in this window.
To change it, press “Disconnect” and confirm: the three choices reappear.
If ChatGPT is not connected yet, use “Pairing code” and complete its connection. The displayed code works once and lasts 10 minutes.
If the AI’s app is missing from the computer, use its download link and try again after installing it.
When a control prepares an AI request, follow the displayed window: with ChatGPT, paste the request into the chat and send it; when the local model opens a prefilled request, press Enter. Preparing text is not the same as publishing it.
Local model, private
A local model runs on your Mac or Windows computer.
Install and start a local runtime such as Ollama or LM Studio, and download a model.
Choose Local model under Your AI; select the runtime and an available model in the panel.
Press “Talk to AI” to open the local chat inside the app.
If no model is available yet, the app shows the setup panel. Local chat and opening Claude or ChatGPT are separate paths.
Disconnect the AI
Disconnect acts on the selected AI; memory stays on your computer.
For Claude, confirm disconnection: OmniAgents removes its connection from Claude’s tools. Close and reopen Claude if requested.
For ChatGPT, disconnection revokes the remote OmniAgents connector. This also applies to other web assistants connected to the same address.
For the local model, disconnection clears the active choice; the model remains installed.
To use an AI again, select it and complete the required connection.