What your agents can do
Agents are the primary users of your brain, so the tools cover the whole life of a document: finding it, reading it, writing it, filing it, keeping the vocabulary tidy, and getting things back after a delete. Twenty-three tools in five groups.Finding and reading
Writing
Organizing
Curating the vocabulary
Deleting and restoring
The tools describe themselves. Each carries its own instructions, and
get_taxonomy hands back your brain’s actual structure plus an example of every document format — so a connected agent discovers what it needs at the moment it needs it, without reading any of this. The agent skills teach the workflow up front, which saves the discovery round trips.
Everything in this documentation is written for you. Your agents do not need it: their instructions arrive with the tools themselves.
Why agent writes stay correct
Every write is checked before it lands, and every rejection says two things: what is wrong, and how to fix it. That is written for the agent, not for you, because the agent is the one who has to act on it. An unknown format lists the valid ones. An ambiguous folder name lists the candidates. A property nobody has defined comes back with the list of ones that exist. A body that is nearly-but-not-quite right comes back corrected, ready to resend. The result is that agents fix themselves inside their own loop rather than leaving half-written documents for a person to find later. A rejected write is a normal step in a working agent loop, not a fault — if you watch your agent work, you will see them, and they are the system doing its job. Three behaviours are worth knowing as the person responsible for the brain:- An agent asked to delete a folder that is not empty is told how much would go, and told to confirm with you before anything is removed.
- When a workspace hits its plan’s document limit, agent writes are blocked with the same explanation you would see. See Plans and limits.
- A locked document refuses every agent write — edit, retag, rename, move, and delete alike. Agents can see the lock before they try, the rejection names who can unlock it, and it explicitly tells the agent not to write a copy elsewhere instead. Locking happens in the web app only; there is no agent tool for it.
Connect an agent
Sign in from your client, or use an access token.
Agent skills
Skills that teach an agent these tools and how to use them well.
Document formats
What each format is for, and what you can do with it.