Agents

Memorywhat it learns, kept where you can see it.

Every agent keeps its own memory: an About itself block it reads every turn, notes it writes as it works, and lessons Wakato picks up from its chats once they go quiet. You can read, edit, undo and refuse anything it remembers; the small AI cost of learning is paid from its home space's wallet.

Every agent has its own memory. It remembers what it learned doing its work: how you like things done, facts about a customer, a mistake not to repeat. The memory belongs to the agent, so it follows it everywhere, to a new engine or a new computer, and it is deleted with the agent.

You can read everything it remembers, change it, and take back anything you do not want it to keep. Learning from its chats uses a small AI model, paid from its home space's wallet.

How it works

About itself

A short block it reads first on every turn: what it is for here, how it works, the standing rules you gave it. You can edit it.

Notes

One note per thing it knows, each with a one-line summary it always sees. It opens the full note when the task needs it.

Past conversations

A summary of each conversation it had. It can look back through them and through its own chats when you ask about something from before.

Two ways it learns

  1. It writes a note itself. While working, an agent can decide something is worth keeping and write it down. A line appears in its chat starting with Noted, followed by the note titles, with Undo.
  2. Wakato reads its chats. When a thread has been quiet for a few minutes, Wakato reads what is new and writes down what is worth keeping, up to a few notes at a time, plus a short summary of the conversation. A line appears in that thread starting with Learned, with Undo.

A line that has been undone reads Taken back.

What it does not keep

There is no shared memory for the whole space. What everyone must know belongs in a document, a table or the agent's instructions, where people can see and edit it. The agent also does not memorise who is on the team; it looks up the space's members when a task needs them.

The Memory tab

Open the agent's page and choose Memory. At the top you see how many notes it has and how much it always loads, for example "12 notes · always loaded ≈ 900 tokens".

  • About itself: click the text to edit it, then save.
  • Learned N things from "thread name": what it picked up in the last few days, each note with Not this.
  • The notes: every note, with who wrote it ("noted by" the agent, "learned" from a chat, or "by you"). Press New note to write one yourself.
  • Read its chats now: reads its chats right away instead of waiting for them to go quiet.

Not this means never again

Pressing Not this on a learned note throws it away for good, and the agent will not learn it again. If you later write that note yourself, your version stands.

Do it

Open its Memory tab

Open the agent's page and choose Memory. In its chat panel you can also use the Memory row.

Check About itself

Read the About itself block. Click it to add a rule, such as "Always write prices in euros", and save.

Review what it learned

Look through the Learned strips. Press Not this on anything wrong or private.

Undo from the chat

If a Noted or Learned line in the chat names something you do not want kept, press Undo on that line. Every note it wrote goes back to how it was.

Teach it on purpose

Press New note and write the fact yourself, or simply tell it in the chat: "Remember that Sam approves every refund."

What your agents can do here

  • Write, update and search their own notes, by words or by meaning.
  • Look back through their own past conversations.
  • Read their About itself block at the start of every turn.

An agent cannot write a note you refused, and it cannot read another agent's memory.

Questions

On this page