Keep multiple AI providers connected at once

You can keep a custom or local model connected at the same time as OpenAI, Anthropic, or Google. Saving a custom endpoint no longer hides your other keys, so you can use Claude for a stakeholder draft and a local model for a sensitive customer transcript in the same session.

Captured context is also easier to trust across a longer conversation. Context you capture from a page, a screenshot, or a text selection now stays attached to the chat across follow-up questions, instead of being cleared the moment you send the message it was captured for.

What's new

• Custom OpenAI-compatible endpoints now appear in the model switcher as another provider, alongside catalog providers you have keys for.
• Saving a catalog key or a custom endpoint still selects that connection, but the others stay available to switch to.
• Onboarding tests every connection you have added, not only the last one saved.
• Captured page, screenshot, or selection context stays attached across follow-up questions instead of clearing after the message it was captured for. Ask a first question about a captured roadmap doc, then a follow-up a few messages later, and the assistant still has it. Remove a context chip yourself when you are done with it.