👤 For pipe admins
🔐 Available on all plans
🎯 For those who want to monitor the performance and accuracy of their AI agents.
The AI Agent Logs area gives you full visibility into how an agent behaves inside your processes. It lets you understand the decisions an agent made, the data and knowledge it used, and where an error occurred, so you can trust and improve your agents without opening a support ticket.
Logs help you:
See what the agent did during each execution, step by step.
Understand why it decided something, in plain business language.
Trace the source of every decision, including which document or field it relied on.
Find errors quickly, including the step where a failure happened and its cause.
Confirm skipped actions, so you know when the agent chose not to act and why.
How to open AI Agent Logs
In the pipe’s top header, click AI Agents.
Open the Logs tab in the upper-left corner. You will see the most recent executions, each showing the agent name, the behavior that was triggered, the card that was processed, and the execution status.
Click See Details and then Tracing to inspect it.
Before you start
To access the logs, you must:
Have read access to the pipe. Any member with read access to the pipe can see the logs.
Have at least one AI Agent configured and active in the pipe.
Step-by-step
1. Access the AI Agents area
Inside your pipe, click the AI Agents button located in the top header.
2. Navigate to the Logs tab
In the upper-left corner of the agents' screen, click the Logs tab. Here you will see a list of all recent executions, including the agent's name, the triggered behavior, the processed card, and the execution status.
3. Analyze the Summary
To understand a specific execution, click the See Details button next to the desired log. The first view is the Summary:
Status: Success, failure, or processing.
Dates: When the execution started and ended.
Context: Which trigger started the agent and which card was affected (including the Card ID).
4. Explore the Tracing
Click the Tracing tab to view the step-by-step execution. This area details the flow followed by your agent and which steps were completed. We are constantly evolving this section to provide even more data and visibility over time regarding your agent’s behavior.
What the Tracing shows
📍 This section is updated as new tracing nodes ship.
Execution Plan
Each step now includes the agent’s reasoning in plain business language, derived from what the agent actually did during the run (not a generic, after-the-fact explanation). You can read, in order, how the agent moved from the trigger to its final action.
If a particular step does not expose reasoning, the log flags that explicitly, so you are never left guessing whether information is missing or simply unavailable.
Knowledge Sources
For agents that use knowledge bases (for example, intelligent document processing, data lookups, or attached documents), the Tracing shows the exact source behind a decision: the document name, the section, and the specific excerpt (chunk) the agent relied on. This lets you verify that the agent consulted the right source and used it correctly.
Agent actions
Each Action appears as its own node in the flow with its own reasoning, planning or execution details, for example: Update card, Move card, and the other actions your agent can run.
Plain Text
When the agent consults a Plain Text knowledge source (static business rules, lists, or guidelines pasted into the agent), Tracing shows that consultation as its own node. You can see that the agent used this source and which content it relied on to support the decision, without opening a database or a document attachment.
Data Lookup
When the agent queries a pipe or database through a Data Lookup knowledge source, Tracing shows that lookup as its own node. You can see the field and condition/value searched, and whether the lookup found a match (with the matched record/sources) or no match.
Notes for this version:
Archived or deleted cards are not shown in the node.
If you do not have permission to see a matched card, the node shows a locked state without the card title.
When the agent uses Web Search, Tracing shows a dedicated node with the search query and, when available, the results (summaries and links). Use it to confirm the agent looked outside the pipe for real-time context and that the search influenced the decision correctly.
Calculation & Analysis (Code Interpreter)
When the agent uses Calculations & Analysis, you can inspect the code that ran in the secure sandbox and the output (shown as a second code block). A single run may include more than one node. Use it to verify exact math, cross-checks, or deterministic validations, and to debug when a calculation or rule did not behave as expected.
When a behavior runs with Max Effort enabled, Tracing shows a node with the final analysis: planning summary, tasks, and results. Related skill nodes (for example Web Search, Code Interpreter, or Data Lookup) appear above it in execution order; plan and action details appear below. The to-do list is available in the final state of the node. Use it to understand the iterative path on complex runs (long documents, cross-document reconciliation, multi-factor rules) instead of a single-shot execution.
Understanding failures and statuses
Logs are essential for fine-tuning your agent's behavior:
Failed: If a step fails, the Tracing shows which step stopped. Some failures also show the cause, such as a missing required field or a permission error.
Skipped: The agent will show when it decided not to perform an action, accompanied by a logical justification for that decision.
Conditions not met: The agent was triggered, but the automation trigger conditions were not met on this card. No action was executed.
Validation checklist
Use the logs to validate a new or edited agent:
Confirm the final Status shows Success.
In Tracing, follow the Execution Plan to check that the reasoning matches what you expected.
Under Knowledge Sources, confirm the agent consulted the correct documents, sections, and fields.
Identify any problematic step and adjust the agent’s instructions accordingly.
Send us feedback
Have a suggestion for AI Agent Logs?
Use the Send feedback button in the top-right corner. Your input helps shape what we build next.