> ## Documentation Index
> Fetch the complete documentation index at: https://docs.pinata.cloud/llms.txt
> Use this file to discover all available pages before exploring further.

# Chat

> Talk to your agent through the web interface

The Chat tab is your main conversation with the agent. Messages stream in real time. You see your prompts, the agent's responses, any tools it called along the way, and — if you enable it — its thinking.

Open your agent → **Chat**.

<Frame>
  <img src="https://mintcdn.com/pinata/p6XBuxeipyyP6Ugl/images/image-30.png?fit=max&auto=format&n=p6XBuxeipyyP6Ugl&q=85&s=cba88b71308ce694f1ac0b19c55ba465" alt="Image" width="1969" height="1358" data-path="images/image-30.png" />
</Frame>

## Sending a message

The composer is at the bottom of the page.

* **+** opens an attachment picker. Files you attach land in `<workspace>/uploads/` inside the container, and the agent gets the path in its context.
* The text input is the message itself. Press Enter to send, Shift+Enter for a newline.
* **MODEL** picks which model handles this turn. It defaults to whatever you set as the agent's default — see [Models](/agents/models).
* The send arrow submits.

## Slash commands

Type `/` at the start of a message to run a command instead of a regular prompt:

| Command   | Effect                                                            |
| --------- | ----------------------------------------------------------------- |
| `/new`    | Starts a fresh session — the agent forgets the prior conversation |
| `/status` | Prints the current model, session info, and key flags             |
| `/help`   | Shows the commands your agent supports                            |

The list depends on your agent's OpenClaw version and any skills you've attached, so `/help` is the best way to see what's actually available.

## Reading agent responses

Each turn the agent sends back has a few elements you'll see in the chat:

* **Thinking block** — collapsed by default, labeled `Thinking ▾`. Click it to expand. Only appears for models that expose reasoning.
* **Message text** — the actual reply. Renders markdown, including code blocks and tables.
* **Tool calls** — if the agent called a tool (a shell command, a web search, a skill), you'll see a labeled block with the arguments and result. This is your single best window into *why* the agent answered the way it did.
* **Model badge** — after the response, a small badge shows which model produced it. Useful when you've been switching models mid-conversation.

When you want to debug a bad answer, expand the thinking block and the tool calls before anything else. That's almost always where the answer is.

## Switching models mid-conversation

Use the **MODEL** dropdown in the composer. You can switch between any models you've enabled for this agent — start a turn with a cheap, fast model, then bump up for a hard question. Each turn is tagged with the model that produced it, so the conversation stays readable.

If you want to confirm what the agent is currently using without scrolling, send `/status`.

## More room for the conversation

Click **Hide navbar** in the top-right corner to collapse the left sidebar — the conversation gets the full width of the screen. Click it again to bring the sidebar back.

## Tips

* Long-running tool calls show a progress indicator. Wait for it before assuming the agent is stuck.
* If the chat looks disconnected and stays that way, the gateway probably needs a kick. Open **Danger** → **Restart Gateway**.
* If you change a secret, attach a new skill, or add a channel, restart the gateway so the agent picks up the change.

For deeper debugging, the [Logs](/agents/logs) tab streams everything OpenClaw is doing under the hood. See [Troubleshooting](/agents/troubleshooting) for common failure modes.