Setting up a new agent with Claos

New agent opens a conversation with Claos, with the agent's settings docked beside it. That is what this article is about. If you would rather fill the form in by hand, the docked panel is that form, described in Building a custom agent.

How it goes

Claos opens by asking what the agent is for, with two chips under its message. Ask me starts a short interview. I'll do it myself leaves you the docked settings panel to fill in by hand, with Claos there if you want a hand with any field. Nothing is saved until you answer or click a chip: leave the page before that and there is no conversation to come back to.

Either way, tell Claos what the agent should do and what to call it, in your own words: "a researcher who watches three competitors, call it Scout". If you have no name in mind, Claos suggests one and you can change it later. That first answer is enough for Claos to start: the agent appears in your agent list, and in the panel on the right you watch Claos fill it in, its cursor moving from the name to the job title and the description, typing each one in. Until then the circle in the panel shows Claos's mark. The agent's icon is one of your Dinodex characters, a starter or one you have hatched, picked so it does not match an agent you already have; change it any time from the icon in the panel.

Then Claos asks how the agent should sound, with a few voices as chips under its message ("Friendly and clear", "Concise and direct"). Pick one or say it your own way, and Claos writes the instructions once, in that voice. After that it asks which tools the agent works in, which skills it should have, and whether it should keep memory of its work, one at a time, the way a colleague would. Keep answers short. If Claos already knows enough, it skips the question. Answering with a chip never changes what is already written: the name, the icon and the instructions change only when you ask for a change.

Each question comes with its answers as chips under it, so you can tap rather than type. Tools come as chips under Claos's question, with See all plugins beside them, which opens the panel's Plugins tab so you can browse everything; the skills question has See all skills the same way. A tool is added to the agent only when you choose it: by its chip, or by naming it in your answer. A tool already connected in your workspace shows as a chip too, and one click adds it. A Connect chip connects that tool on the spot, and a Request access chip asks for one that is not open yet. You can also name any other tool, or press Carry on with built-in tools: web search and browsing, pages and tasks need no connection at all.

Skills come as chips too. They start picked, so untick the ones you do not want. To add one the list does not have, type it in the Add your own skill box beside the chips and press Enter: it becomes a chip of its own. Then press Add. Claos adds the ones from the list, and writes each of yours, adds it to your workspace's skill library and gives it to the agent. On the right, the cursor clicks over to the Skills tab and the skills arrive one at a time.

Memory is a choice of two chips, Keep memory or Start fresh each time. Claos switches it on the Memory tab, then the cursor clicks back to Instructions, where the whole agent is easiest to read.

When the last answer is in, Claos says the agent is set up and asks whether you would like anything changed, with a Done chip under its message. Say a change in the chat ("make it more formal", "drop the research skill") and you see it made in the panel. You can also edit any field in the panel by hand at any point. Press Done when you are happy: setup closes and a chat with your new agent opens. The setup conversation stays in the agent's conversations, and picking Claos in the agent list afterwards starts a fresh chat with Claos rather than reopening it.

Until a name is agreed no agent is made, so leaving before that leaves no agent behind. After the name, the agent exists in your list and you can come back to the conversation to carry on. To leave setup at any time, close the panel on the right or pick any agent in the list.

After it is made

The agent is an ordinary agent. Its page holds the same settings tabs as any other, its plugins connect the same way (see Connecting an integration), and you can put it on a task or chat with it right away. What you agreed is there to read in the setup conversation.