The chats
You can also open a chat scoped to a project, dashboard, or a single app widget; it carries that context into the conversation.
How a chat uses tools
The assistant doesn’t answer from thin air. It calls tools—small, named operations like “list the entities,” “search the databases,” or “add a field to this entity.” It chains several tool calls to get your answer or make your change. You can watch this happen. Each assistant message has a Netter is thinking… line you can expand to see the timeline: its reasoning interleaved with the tool calls it made, each shown with an icon, a name (list_entities, Edit, describe_entity), and what it acted on. Nothing is hidden—the tool activity is right there in the transcript.
Tools respect your permissions
This is the important part: the assistant acts as you, not above you.- Tools that return data—answers, charts, search results—apply your resource access and row-level security. The assistant can’t surface a row, dataset, or entity you couldn’t open yourself.
- Tools that change things—creating an entity, editing fields, promoting a database—check the same permission levels the equivalent buttons do. If you lack the level, the tool refuses and the assistant tells you why instead of doing it anyway.
Next
- Chat tools — the complete tool list and the permission each one respects.
- Permissions — the access model the tools enforce.
- Build an app — design a viewer surface through the app chat.