Specialized agent logs
View and download execution logs for your specialized agents in Mark (Optimizely Agent Platform). Logs record each execution's inputs, outputs, steps, timing, and credit usage. Use them to debug your agent, monitor performance, and evaluate output quality.
Use the Agent activity logs tab to see execution activity across all agent types.
Access your agents
- Go to home.optimizely.com.
- Select your organization.
- Click Mark.
- Click Agents.
- Select the Your Agents tab.
View specialized agent logs
To view executions across all specialized and workflow agents at once, use the Agent Activity Logs tab on the Agents page. See Agent activity logs for details.
To view logs for a single specialized agent, complete the following:
- Follow the steps in the Access your agents section.
- Click the agent for which you want to view its logs or click More > Edit Agent.
- Select the Logs tab.
- Click an execution to view its logs.
Specialized agent logs overview
Started and running
The specialized agent logs for executions still running display information like the following:
- Agent Name – The name of the agent invoked.
- Agent ID – The unique identifier for an agent, prefixed with @.
- Close – Close the execution details.
- Execution tab – Contains the agent's execution details.
- Agent Details tab – Contains the agent's configuration details.
- View Agent – Open the agent's configuration page.
- Version information – The version number and timestamp of when the agent was started.
- Status – Displays Started as the agent is running.
- Stop Execution – Click to stop the agent.
- Agent Name – Name of the agent. Click Copy to copy the name to your clipboard.
- Execution ID – The unique ID of the agent's execution. Use this ID to identify a specific agent run. Click Copy to copy the ID to your clipboard. See the Access a specialized agent's execution ID section for information.
- Execution Time – The timestamp of when the agent was started.
- Triggered By – The name of the user who started the agent.
- CWD – The current working directory (CWD) that the agent used for the execution.
- Input – The variables that were used in the execution. Click Copy to copy the variables to your clipboard to reference and debug agent runs.
- Output – The agent response from Mark including any error messages. The output updates as the agent runs.
- Execution memory – Individual steps the agent took during the execution.
Completed and failed
The specialized agent logs display the following information for Completed and Failed executions:
- Agent Name – The name of the agent invoked.
- Agent ID – The unique identifier for an agent, prefixed with @.
- Close – Close the execution details.
- Execution tab – Contains the agent's execution details.
- Agent Details tab – Contains the agent's configuration details.
- View Agent – Open the agent's configuration page.
- Version information – The version number and timestamp of when the agent was started.
- Status – If the agent is Completed or Failed.
- Open in Test Run – Open the agent in a Test Run with the same variables entered.
- Agent Name – The name of the agent invoked. Click Copy to copy the name to your clipboard.
- Execution ID – The unique ID of the agent's execution. Use this ID to identify a specific agent run. Click Copy to copy the ID to your clipboard. See the Access a specialized agent's execution ID section for information.
- Execution Time – The timestamp of when the agent was started.
- Triggered By – The name of the user who started the agent.
- CWD – The current working directory (CWD) that the agent used for the execution.
- Duration – The total time the agent took to run.
- Guardrail Status – If the agent passed or failed the guardrails. See Configure Execution Guardrails.
- Credits – Approximate credit usage for this execution. Final usage may vary slightly. See Mark credits.
- Impact – The impact that the run recorded, if the agent has an impact estimate. See Agent impact reporting.
- Input – The variables that were used in the execution. Click Copy to copy the variables to your clipboard to reference and debug agent runs.
- (Optional) Artifacts – A link to the canvas output if one was created during the execution's run.
- Output – The agent response from Mark including any error messages. Click rerun to run the agent again.
- Execution Memory – The steps Mark took during the execution. For multi-turn agents, Mark organizes memory by turn, with each turn displaying the user input and agent response. For single-shot agents, memory shows the steps Mark took to form its response, including tool calls and intermediate outputs.
If you added preferred output examples for the agent, the logs display an Evaluation Score. The score rates the quality of the execution and includes a Pass or Fail indicator.
Export specialized agent execution logs
Download a record of your specialized agent's execution history as a comma-separated values (CSV) or Excel file. Use it to track usage, review inputs and outputs, and measure return on investment (ROI).
- Follow the steps in the Access your agents section.
- Click the agent for which you want to view its logs or click More > Edit Agent.
- Click More > Export logs.
- Select the columns you want to include. See Available columns. To clear all selections, click Deselect all.
- Select your format, CSV or Excel, from the Format drop-down list. See Export formats.
- Click Export.
Available columns
The following column groups are available for export:
Execution info
- Agent Name – The name of the agent.
- Execution ID – The unique ID of the execution. Use this ID to identify a specific agent run.
- Execution Time – The timestamp of when the execution started.
- Triggered By – The name of the person who ran the agent.
- Version – The version of the agent that ran.
Input and output
- Variables – The input data that was passed to the agent.
- Output – Mark's response, including any error messages.
Performance
- Status – Whether the execution completed or failed. See Execution statuses.
- Duration – Total time the execution took to run.
- Evaluation Score – The quality score for the agent execution.
- Credits – Approximate credit usage for the execution. See Mark credits.
Impact
Select the Impact checkbox to add the following columns. See Agent impact reporting.
- Impact – The impact recorded for the run. In Excel exports, this column holds numbers that you can sum.
- Impact Unit – The unit of measurement for the impact.
- Impact Category – The category of the impact.
Export formats
Choose from the following export formats:
- CSV – Plain text file that opens in any spreadsheet tool. Best for importing into other systems or doing programmatic analysis.
- Excel – Formatted workbook with bold headers and a text-wrapped output column. Best for handling long outputs gracefully.
Export limits and file details
The following limits and naming conventions apply to exported files:
- Exports include up to 500 of the most recent executions. If your agent has more, the page displays the total count it exports.
- The file name matches your agent's name. For example, My_Agent-executions.csv.
- The Output column displays the error message for failed executions.
- The Triggered By column displays the full name of the user who ran the agent.
Set a specialized agent execution as an evaluation sample
Set a specific specialized agent execution as an evaluation sample for Mark to evaluate its responses against. Mark uses the evaluation sample output to create an Evaluation Score. Add up to five eval samples outputs. Include at least three diverse outputs for the best results.
Follow these steps to set an execution as an eval sample:
- Follow the steps in the Access your agents section.
- Click your specialized agent or click More > Edit Agent.
- Select the Logs tab.
- Click More for an execution and select Link as Eval Sample.
Linking as an eval sample automatically adds the execution to the specialized agent's preferred outputs. After configuring your preferred outputs, ensure you set the baseline quality score. See Add Output Evals for steps and information on what different scores mean.
Access a specialized agent's execution ID
The Execution ID is a unique identifier for a specific agent run. Use it to reference a run in Optimizely Support requests, share it with collaborators, or monitor a long-running execution. Use the Execution ID for any of the following:
- Tracking and monitoring – Track the progress and status of a long-running agent execution.
- Retrieving results – Use the Execution ID to retrieve results when an asynchronous execution finishes.
- Debugging and auditing – Trace a specific run and review what happened. This is especially useful when diagnosing issues or unexpected outcomes.
- Sharing and support – Reference a specific agent run when working with support teams or collaborators.
Follow these steps to access an agent's Execution ID:
- Follow the steps in the View specialized agent logs section.
- Click the run whose execution ID you want to retrieve.
- Click Copy for the Execution ID.
Ask Mark about an execution
Use Mark chat to ask questions about a specialized agent execution in natural language. Chat helps you understand what happened during a run without reading through every memory chunk and tool call manually.
Start a chat-based debugging session using one of the following methods:
- Open Mark chat from the execution log details view to load the execution context automatically.
- Open Mark chat directly and reference the execution ID in your prompt (for example, Debug execution abc-123). See Access a specialized agent's execution ID for how to find the execution ID.
After Mark loads the context, ask questions such as the following:
- Why did the agent call this tool?
- What memory chunks were retrieved on turn 3?
- Why did this execution fail?
- What was the total token usage?
- Show me the full tool call sequence.
Mark restricts execution debugging to your instance.
Note
If you use Opti ID, administrators can turn off generative AI in the Opti ID Admin Center. See Turn generative AI off across Optimizely applications.