> ## 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.

# Logs

> Stream OpenClaw's output in real time

The Logs tab is a live stream of what OpenClaw is doing inside your agent. New entries appear without refreshing — the tab uses a WebSocket under the hood. If the small status badge in the header reads `WEBSOCKET CONNECTED`, you're streaming. If it reads `WEBSOCKET NOT CONNECTED`, either the agent isn't running or the connection dropped — hit the refresh button or check the agent's status.

<Frame>
  <img src="https://mintcdn.com/pinata/p6XBuxeipyyP6Ugl/images/image-38.png?fit=max&auto=format&n=p6XBuxeipyyP6Ugl&q=85&s=8f9301e35c6a64034bcd38ad884426bb" alt="Image" width="1969" height="1358" data-path="images/image-38.png" />
</Frame>

## Reading the stream

Each log line has four parts:

* **Timestamp** — when it happened
* **Level** — severity (`TRACE`, `DEBUG`, `INFO`, `WARN`, `ERROR`, `FATAL`)
* **Source** — which component generated the line (e.g. `gateway/reload`, `cron/runner`)
* **Message** — the content

Click the **+** on the left of any row to expand it. Useful for stack traces and structured fields that don't fit on a single line.

## Filtering

The six colored checkboxes — TRACE, DEBUG, INFO, WARN, ERROR, FATAL — toggle log levels on and off. When you're hunting a problem, uncheck everything except WARN and above to cut the noise.

The **Search logs** box does a substring match against the message. Combine with level filters: "show me ERROR lines mentioning `telegram`."

The **AUTO-FOLLOW** checkbox keeps the view pinned to the latest entry. Uncheck it when you want to scroll back without the view jumping.

## Exporting

**EXPORT VISIBLE** downloads whatever the current filters are showing — useful for sharing with support or filing an issue. It's only what you see; the export respects your filters.

## Where the logs live on disk

Inside the container, OpenClaw writes daily log files to `/tmp/openclaw/openclaw-{date}.log`. From the [Console](/agents/console):

```bash theme={null}
ls /tmp/openclaw/
tail -f /tmp/openclaw/openclaw-$(date +%Y-%m-%d).log
```

That's identical to what the Logs tab streams, with one difference: only the last 100 lines are retained when the gateway restarts. If you need long-term retention, export periodically.

## Reading logs over the API

For scripts and incident response:

```bash theme={null}
curl -H "Authorization: Bearer $PINATA_JWT" \
  https://agents.pinata.cloud/v0/agents/$AGENT_ID/logs
```

Returns the latest 100 lines as a single string. For full streaming, the UI uses a WebSocket on the agent's subdomain — see [API → Streaming logs](/agents/api#streaming-logs).

## Common sources, and what they tell you

| Source                         | What it means                                                                           |
| ------------------------------ | --------------------------------------------------------------------------------------- |
| `gateway/reload`               | Gateway reloaded its configuration — usually after a secret, skill, or channel change   |
| `gateway/ws`                   | WebSocket lifecycle (connect, disconnect, reconnect)                                    |
| `cron/runner`                  | A scheduled task fired — which one, and what happened                                   |
| `agent/turn`                   | One conversation turn — request in, response out                                        |
| `skill/<name>`                 | Output from a specific skill                                                            |
| `script/build`, `script/start` | Lifecycle script output (also tee'd to `/tmp/user-build.log` and `/tmp/user-start.log`) |

If something's broken, start at ERROR. If you see ERROR lines and can't make sense of them, [Troubleshooting](/agents/troubleshooting) maps symptoms to fixes, and the [Error reference](/agents/errors) covers common API error responses.