Organizer
Creates folders and tags, moves items into folders, assigns and removes tags, and renames folders and tags.
AI agents
A supervisor plans your request and delegates to three specialists: Organizer, Writer, and Researcher. They act on your workspace, and you watch the work happen.
Each is a real independent agent with its own toolset and system prompt. Delegation is one level deep, inside a single turn.
Creates folders and tags, moves items into folders, assigns and removes tags, and renames folders and tags.
Creates docs, notes, and bookmarks, renames, archives, pins, trashes and restores, and files new content into a named folder.
Lists, reads, and searches uploaded files, gets a file outline, and fetches a web page.
The Chat agent searches the workspace, reads a specific doc, delegates, and handles destructive deletes.
Every action is typed and declares its risk tier up front: read, write, or destructive.
The current split is 11 read, 17 write, and 3 destructive.
Permanent delete, delete folder, and delete tag stop and show an Approve or Reject card first.
Delegation steps stream live, and on a reasoning model the thinking streams too.
An approval request waits for you. Answer it later, even from another device.
Stop re-prompting the same job.
Define your own agent, its instructions, and the exact tools it may use.
Name a workflow once and invoke it later.
Hand a long job to an agent and carry on working while it runs.
Run an agent on a timer, a trigger, or a webhook: a daily digest, a weekly summary, a tidy-up.
What happens inside one turn.
Describe the job in plain language in a chat.
The supervisor plans the request and picks specialists.
Specialists read, write, and file, and you watch each step.
Destructive steps wait for your Approve or Reject.
Only destructive actions are gated: permanent delete, delete folder, and delete tag. Trash, archive, rename, move, and tag changes run without a prompt, and most are reversible.
An interactive turn is bounded: the supervisor runs up to 12 steps, Organizer and Writer 8 each, and Researcher 20. For longer work, use a background run.
Yes, through schedules, webhooks, and background runs. Finished work and mentions collect in the inbox.
No. Your chosen model drives the supervisor. Each specialist runs on its own assigned model suited to its job.
Agents that act on your workspace, with you as the last word on anything destructive.