Workspace AI providers and models
Set up AI models, credentials and connections without confusing their roles.
Vault and connections
In Settings → Vault and connections, review available providers and add the connection you want to use. Choose a model for the right purpose: conversation, speech recognition and voice output may need different services. You can connect a provider such as OpenRouter or use a local model, depending on what you want for each function.
Check a connection
- Obtain credentials from your chosen provider.
- Add them in the corresponding panel, without pasting them into chat.
- Select the model and its supported capabilities.
- Use the available connection check and try a short request.
- Review limits and usage in your provider account.
An app subscription and an external provider’s usage are separate services. Providers may have their own costs, limits or requirements. Custom models can only perform capabilities they actually support.
What uses what
| Function | Where it is used |
|---|---|
| Conversation | Chat, workspace drafts and Internet search |
| Voice and recognition | Voice assistant and dictation |
| Omnibar AI | Omnibar: ideas based on context and creating actions with AI |
These connections are not needed to create and launch workspaces manually. For failed AI requests, check the plan, connection and model capability first; see troubleshooting.
External agents
This page is about the AI services the app uses. If you want the opposite, for your agent (Claude Code, Codex, Antigravity…) to use Workspace Launcher from its chat, see connect your AI agent.
Reviewed on