Features
An assistant in the room, not in a side tab.
It reads your content model, drafts and rewrites copy, and can take a whole task. Anything that changes your content stops and waits for a human first.
The assistant docked beside the work, stopping on an approval card before it changes anything.
How it works.
- In the same session
- The assistant docks beside your work and keeps the conversation as you move around the app, so context does not reset every time you open a different entry.
- Approve before it writes
- Reading the model, listing entries and searching happen freely. Creating a collection, adding a field or changing an entry stops on a card showing the exact change, and waits for you to allow it.
- It works in the open
- Every change it proposes is shown as the exact write it wants to make, on the entry and field it affects, so there is nothing to diff by eye after the fact.
- It knows your content
- Search runs across your entries, so answers come from what you have actually written rather than from a general impression of your industry.
- Delegate the whole task
- Give it a job rather than a prompt: add the field, fill it in across the drafts, and report back. You stay in the loop at every write.
- Or bring your own agent
- The same tools are served over MCP, so Claude, Cursor or whatever you already code in can read the model, search entries and write bodies. It signs in as you over OAuth, there is no key to paste, and your client is what confirms each write.
Who this matters to.
The same feature, described from the side of the gap you happen to be standing on.