A Billion Agents User guide
Contents

Agents

Agents and Agent info

The list of your agents, delegates, and Agent info, with an agent’s knowledge, its session and Fresh start.

An agent is an AI that works for you, with a job, a name and files of its own, for as long as you keep it. It works when you message it, and is idle otherwise.

The list of agents

The Agents tab lists every agent you have, with its picture, its name, what it is for, and how many conversations it has, or “working” in green while it works.

The Agents tab: New agent at the top, then “7 agents · 0 working”, and agents such as Bill Payer, with its delegate Receipt Sorter indented under it, Chief of Staff and Researcher.

Agents you created are listed by name. Under each, indented, are its delegates: agents that it created itself for a piece of work, which do that work on their own and report back to it. Search agents finds an agent by name. New agent at the top starts a new one (see starting a new agent).

Click an agent to open its Agent info.

Agent info

Agent info is about one agent. You also reach it from a conversation’s Conversation info.

Agent info for Bill Payer: its time worked this month, “When you message it” under Wakes up when, and its knowledge files, such as knowledge/payables/recurring.md with its first line.

At the top are the agent’s picture, name and handle (such as @bill-payer), and who it reports to: “Top-level agent · reports to you” for an agent you created, or “Delegate of” the agent that created it. Two buttons follow:

  • New conversation starts a new conversation with this agent.
  • Inspect opens the inspector, a deeper view of how the agent works.

While the agent works, a line says where, such as “Working in Plan the team offsite”, with Stop to stop its reply.

Then, in order:

  • Purpose: what the agent is for.
  • Conversations: its four latest conversations. When it has more, Show all lists them in the left pane.
  • Time worked: how long it has worked this month.
  • Wakes up when: what makes it start work. Today that is when you message it, and for a delegate also when the agent that created it hands it work.
  • Can: its skills, written how-tos it reads when a task needs them, when it has any.
  • Knowledge: the files where it keeps what it learns, each with its first line and when it last changed. Its knowledge carries into every session.
  • Delegates: the agents it created, when it has any.
  • Saved sign-ins, GitHub account and Computer environment: see An agent’s browser, Connecting GitHub and Computer environment.
  • Session, at the bottom.

Sessions and working memory

An agent works in one session at a time: its working session, one stretch of work that all your conversations with it feed. What you say in one conversation, it knows in the next. When a session has run its course, the agent ends it and leaves itself a note to start the next one from.

Session, at the bottom of Agent info, says when its current session started, or “Not in a session”. Click it to see more.

Agent info’s Session section open: “1 conversation · last worked just now”, Working memory at 20% with a bar in three parts, and the Fresh start button.
  • How many conversations the session holds, and when the agent last worked.
  • Working memory: how full the session is, as a percentage, with a bar in three parts: its instructions and workspace, the messages from every conversation, and its own work, like using its tools. Agents usually end a session long before it fills. If it does fill, the agent wraps up and starts a new session.
  • While the agent’s computer is on, a line such as “Its computer is running, since today at 9:14 AM.”
  • Fresh start ends the session, so that the agent’s next message starts with a clear working memory. Click it, then Fresh start again to confirm, or Cancel. If the agent is working, it stops; then it wraps up and ends the session.

With no session open, it says that the next message to the agent starts one.