Chatting with an agent

Chatting is the simplest way to work with an agent: open it, ask, and watch the work happen.

Opening a chat

Open Agents in the sidebar. It takes you back to the last conversation you had. The page has three parts:

• On the left, your agents. Claos, the built-in assistant, is first, then your own agents, most recently used first, each with the last thing it said. Search at the top finds an agent by name or Job Title.
• In the middle, the conversation.
• On the right, past conversations with the selected agent. Filter them with All, Chats and Runs, or switch to All Agents to see everyone's.

Pick an agent on the left to switch to it. You land in your latest conversation with it; to start fresh, use the new chat button at the top of the conversation.

The chat bubble in the corner of other pages is a quick chat with Claos, with the help articles and the getting started guide beside it (see Getting help inside Claos). Open full chat moves that conversation onto the Agents page.

Sending a message

Type in the box at the bottom and press Enter to send. Shift + Enter starts a new line. On a phone, Return starts a new line and the Send button sends.

When a conversation is new, a few suggested prompts sit under the box. Click one to send it.

Attaching files

Use the paperclip, or drag a file into the box. The agent reads the file itself:

• Documents: PDF, Word, PowerPoint, text, Markdown, JSON and XML. It can answer about a page deep in a long report, or a scanned PDF.
• Spreadsheets: Excel and CSV. The agent calculates over the data rather than guessing.
• Images: PNG, JPEG, WebP, GIF and iPhone photos (HEIC).

It remembers files from earlier in the same conversation, so you can ask follow-up questions. A file can be up to 50 MB. A document over about 300 pages is too long for one message: Claos tells you which file and puts it back in the box so you can split it.

Asking another agent

In a chat with Claos, type @ and pick an agent to have that agent answer this one message, without leaving the conversation. A chat with one of your own agents is with that agent only, so it has no @. To work with several agents at once, use a room.

What you see while it works

The agent's reply streams in as it's written. While it works, a short line shows what it's doing, such as searching your workspace or opening a page. The Stop button ends it at its next step, and what it had produced is kept.

The agent keeps working if you reload the page or your connection drops. The reply picks up where it is when the page comes back.

What a chat can produce

• A reply. Most questions end here.
• A page. When the agent writes a page, it opens in a panel beside the chat as it's written and is saved in the Knowledge Base, in the Artifacts folder unless you've asked for somewhere else. You can edit it right there.
• Tasks. Tasks the agent creates appear as chips in the reply. Click one to open the task beside the chat.
• Images. Ask for a picture and it appears in the reply. See Making images with AI.
• Browsing. If the agent needs a website, the browser opens beside the chat. See The agent's browser.
• Designs. Ask for a social post, slides or a one-pager and the agent makes a first version straight away. It appears as a card under the reply and opens in a panel beside the chat. See Designs.
• Suggested plugins. If the work needs an app the agent can't reach yet, the reply suggests the apps that fit, and you can connect one right there. See Connecting an integration.

A reply that produced something ends with a small bar naming it, which reopens the page or task.

Where past chats live

Every conversation is kept in the list on the right, with the agent's runs on tasks and schedules beside them. Each conversation has its own web address, so you can copy the link from your browser to come back to it. A new chat names itself from your first message; to pin, rename or delete one, use its menu in the list.

Tips

• You can't edit a message after sending it. If a reply isn't quite right, say what to change; the agent keeps the earlier turns in mind.
• A file you attach is for that conversation. For guidance the agent should always follow, put it in its instructions, or in a skill if several agents should share it.
• The settings button at the top of the conversation opens the agent's instructions, plugins, skills and memory without leaving the chat.

Related

• What is an AI agent?
• Putting an agent on a task
• Reading agent runs
• The Knowledge Base