Viewing Logs
- How the Log Level setting on each agent controls what gets logged
- Where to find logs in AutoTalk
- How to use logs to troubleshoot agent behavior
Logs provide a detailed record of your AI agent's activity. They capture messages processed, tool calls made, responses generated, and errors encountered. Reviewing logs is the fastest way to understand why an agent behaved a certain way and to diagnose problems.
How log levels work
Every agent has a Log Level setting, located on the agent's Advanced tab inside the Options group (Name and Agent Type live on the General tab). The log level controls how much detail is recorded:
- Lower verbosity: Only significant events (errors, warnings) are logged. Suitable for production agents that are running smoothly.
- Higher verbosity: Detailed information about every step of processing is logged, including input/output data, tool call parameters, and internal decisions. Ideal during setup, testing, and debugging.
Set a higher log level while you are building and testing your agent. Once the agent is stable and handling real customer conversations, reduce the log level to save storage and make it easier to spot genuine issues.
Where to find logs
Agent logs are available in AutoTalk:
- In the sidebar, open Company → Logs (owner-only).
- You will see a list of log entries from across your agents and system activity.
- Filter or search to find entries related to a specific agent.
What logs contain
Each log entry includes:
| Field | Description |
|---|---|
| Timestamp | When the event occurred |
| Agent name | Which agent generated the entry |
| Event type | The kind of event (message received, response sent, tool call, error, etc.) |
| Input data | The customer message or trigger that started the action |
| Output data | The agent's response or the result returned by a tool |
| Tool call details | Which tool was called, what parameters were sent, and what the tool returned |
| Errors | Any errors or exceptions encountered during processing |
Using logs for troubleshooting
Here are common scenarios where logs help you diagnose and fix problems:
Agent gave a wrong or unexpected answer
Check the log entry for that conversation turn. Look at:
- The system messages that were sent to the model (were they complete and accurate?)
- The conversation history that was included (was relevant context present?)
- The model's raw response (did it misunderstand the question?)
A tool call failed
Find the log entry for the tool call. Check:
- The parameters the agent sent to the tool (were they correct?)
- The error message returned (does it indicate a configuration issue, authentication problem, or data error?)
- Whether the tool was called at all (if not, the tool description may need to be clearer)
Responses are slow
Compare timestamps between the message-received event and the response-sent event. If there is a large gap:
- Check whether a tool call took a long time to complete
- Consider whether the model itself was slow (this can happen with larger models under heavy load)
- Review whether pre-actions are adding unnecessary delay
Agent is not responding
Check whether any log entries exist for the conversation at all. If there are none:
- Verify the agent is assigned to the correct channel
- Check that the bot toggle is enabled for the conversation
- Make sure the conversation is not in a bot cooldown period (agents have no separate status/active toggle — whether an agent replies is governed by channel assignment and the per-conversation bot toggle/cooldown)
- An unanswered attachment: Check React to file attachments on the Advanced tab for additional file handling. Transcribed audio and supported images with image processing enabled are exceptions: they are handled before this flag. Check transcription/vision settings too. See Configuring LLM Models.
Make it a habit to review logs after deploying any changes to an agent's system messages, tools, or pre-actions. A quick scan of the first few interactions will tell you if everything is working as expected.