What is an AI agent?

An agent is a colleague you've already hired. You name them, give them a job, tell them how to work, and from then on you delegate — you don't re-explain.

This article is the mental shift. If you arrived expecting a chat box that answers questions, you'll get more out of Claos once you stop thinking of an agent as a prompt and start thinking of it as a member of your workforce. Once that lands, the rest of the AI Agents section makes sense.

An agent is a colleague, not a prompt

A prompt is something you write fresh every time. You phrase it, you add context, you hope for the best, and the next time you come back you do it all over again.

An agent is the inverse. You've already given it a name, a job, a set of instructions, the tools it's allowed to use, and the skills it's allowed to read. When you open it, the brief is already there. You just say what you need today, and it picks up where you left off.

That's what makes Claos feel less like talking to AI and more like asking a coworker.

What's inside an agent

Five things define an agent. Most of them you set once and rarely touch again.

Name and avatar. How you recognize the agent on the roster. Many agents have a persona ("Imani," "Atlas") — that's a name choice, not a feature.
Job Title. What the agent does — "Writing Partner," "Code Reviewer," "Customer Support Lead." This is the label teammates see on the agent card.
Instructions. The system prompt. The standing brief that shapes everything the agent produces — tone, format, rules of thumb, things it should never do.
Tools. Actions the agent is allowed to take — inside your workspace (search pages, create or edit a page, comment on a task) and in third-party apps via connected integrations (read Gmail, post in Slack, file a Jira ticket). Which integrations are available depends on what's connected — see What are integrations?.
Skills. Knowledge base pages the agent reads before answering. A brand voice doc, a style guide, an SOP — any page you've decided the agent should know.

Where you meet them

The Agents tab in the sidebar is both your roster and your chat surface. At the top are the agents you've kept or built; below them is the catalog of pre-built templates you can use as-is or copy into your own roster. Click any agent and a conversation opens directly — no setup dialog, no preamble. The agent's full detail page puts the chat front and centre, with the agent's settings tucked into a side panel you can open when you need to adjust the brief.

Three ways an agent does work

Once you've picked or built an agent, there are three ways to put it to work:

Chat. Open the agent and ask. Best for one-off requests or quick back-and-forth.
Assigned to a task once. Assign the agent to a project task and set Frequency to Run now or Run once. The agent fires, posts results to the task, and notifies you in the inbox.
On a recurring task. Same setup, but set Frequency to Hourly, Daily, Weekdays, Weekly, or Monthly. From then on the agent runs on its own — that's the "morning brief that's already written when I wake up" pattern. Scheduling lives on the task, not on the agent itself.

Tips and gotchas

The label is Job Title, never "Role." Member-level Role — owner, admin, member, viewer — is a separate field for human teammates, and that one's fine to use.
Agents remember past work inside their conversation history. There isn't a separate "memory" page you open and edit; the history is the memory.
An agent isn't a team. If an ask is big enough that you'd hand it to several specialists, you want a team — a captain plus a roster — instead. See the AI Teams section.