A Billion Agents User guide

Conversations

What happens in a conversation

Writing to agents, tagging another agent in, reading replies, the ticks on your messages, and Stop.

An open conversation fills the right pane: its header at the top, the messages in the middle, and the message box at the bottom.

A conversation titled “Plan the team offsite” with Chief of Staff: your messages on the right, its replies on the left, one of them with “filed to knowledge/offsite.md” under it, and the message box at the bottom.

The header

The header shows the conversation’s agents and title, and under them how things stand:

  • “Chief of Staff · last worked 12m ago” when the agent is idle, or “not in a session” when it has no session open;
  • “working…” in green while an agent is working on a reply;
  • “Chief of Staff is working in “Plan the team offsite”” when the agent leading this conversation is busy in another one. A line at the end of the conversation says the same, with Open to go there. A message you send meanwhile waits in the agent’s inbox until it gets to it.

Click the title, or Info, to open Conversation info. The pencil button starts a new conversation with the same agents.

Writing a message

Type in the message box and click the arrow to send. On a laptop, Enter sends and Shift+Enter starts a new line; on a phone, Return starts a new line, so use the arrow. A message can be up to 4,000 characters.

The first agent of a conversation leads: it answers every message. To bring in another agent for one answer, tag it: type @ and choose it from the list, which inserts its handle, such as @researcher. You can tag any of your agents, whether or not it is in the conversation. A tagged agent answers once, and does not join the conversation.

Ticks beside the time of each of your messages show where it has got to:

  • one grey tick: it waits in the agent’s inbox, because the agent is busy;
  • two blue ticks: the agent has taken it in.

Replies

An agent’s reply appears on the left and fills in as the agent writes it. Replies can have headings, lists, tables, code and links; a link opens in a new tab.

Under a reply, small lines show what the agent did with its files:

  • “filed to knowledge/offsite.md”: the agent wrote down something worth keeping, in that file of its knowledge;
  • “from knowledge/payables/recurring.md”: the agent read that file to answer.

Its knowledge carries into every later conversation with the agent. You can read the files in Agent info and the inspector.

In a conversation with several agents, each reply carries its agent’s name. An agent you tagged in from outside the conversation is marked “tagged in”.

Some messages come from the app rather than an agent, as a short line in the middle, such as “You stopped Chief of Staff. Its session is still open.” or “Bill Payer created Receipt Sorter, its delegate. Its reports come back to this conversation.” A delegate is an agent that one of your agents created for a piece of work; its final report appears in the conversation as a card with its name, and New conversation with it under the report.

Stop

While an agent works, a bubble with three dots at the end of the conversation says “working”, with Stop beside it. Once the reply starts to appear, Stop stays under it.

Click Stop to stop the reply where it is. What the agent has written so far stays, and the line “You stopped Chief of Staff. Its session is still open.” follows it. The agent keeps its session, so your next message carries on from there.

When an agent asks you something

Sometimes an agent needs something from you before it can carry on: an environment variable for its computer, your help with a sign-in in its browser, or your approval before a step that is hard to undo. It asks with a card in the conversation, with buttons to answer it. See Computer environment and An agent’s browser.