Set up AI and an extension

This walkthrough uses the upcoming AI & extensions interface. It requires a Grid desktop Pro build on macOS or Linux containing that interface and an internet connection.

1. Open File → AI & extensions → AI providers. Choose Add API key under your provider, enter your own key, and choose Save and use. Choose Test connection & find models, select a model if desired, and return to the Grid Agent to make a request. A saved key alone does not validate billing or provider access.
2. Open Local models, choose Qwen, then View on a published release. Check its download size, storage requirements, and compatibility. Choose Install only if you want those model weights downloaded onto this computer. For pictures or captured video frames in 0.66.8, choose Help me choose for → Images and documents, then Find a good fit for this computer. Review the Text and images label in its details before installing.
3. Close and reopen the window during the download. The installation continues and its status is restored. After completion, find the exact model under Installed. Choose Try it for a real response and elapsed time, Use this model to select it for the Agent and workbook AI, or Insert in selected cell to use the same model and prompt in your workbook. Pause/resume a large model download to check recovery without discarding its verified partial data.
4. Choose Extensions, search for a published library, open it, and choose your workbook's folder. Install verifies and adds the release to that workspace. Review permissions & activate keeps that folder selected while you review permissions and activate the extension.
5. If your organization has a registry, open Registries → Another registry. Paste its administrator-provided setup link, or enter its name, address, and verification code. Verify and use registry must succeed before it replaces the default. Return to Grid Registry and save to switch back.

If the model does not fit, select a smaller release. If a registry is unreachable or its verification code does not match, fix those settings before retrying. Preview catalog entries and incompatible models never enable Install.

If you already run a model server, use AI providers → Model server → Connect server instead of adding a hosted-provider key. Copy the server's API address, add its key if needed, choose Connect & find models, then Save and use server. Reopen setup to see the saved address and model. You can switch to a hosted provider and return with Use server. The saved selection also supplies new workbook AI_PROMPT and AI_ASK requests. Check the shared-connection status in setup; if the runtime was unavailable when you saved, choose Retry workbook connection. Existing PREDICT formulas keep their explicit model choices.