AI Agent Runs
Every time an AI agent runs, from an AI Agent node or from the API, MonoChat records the run: which harness and models it used, what it was asked, every model and tool call it made, how much of its budget it used, and how it ended.
You find the runs in Settings › Automation › AI Agent Runs. The menu item is shown to users with the Read AI agent runs in automation permission.
The runs table
| Column | Shows |
|---|---|
| AI Harness | The harness that ran the agent. Hover to see the harness type. |
| Status | Queued, Running, Completed, Failed, Budget exceeded, Timed out or Cancelled. Hover a failed run to see its error. |
| Model Calls | Model calls used / Max Model Calls of the budget. |
| Tool Calls | Tool calls used / Max Tool Calls of the budget. |
| Tokens | Total tokens. Hover to see input and output tokens, and cache reads and writes when the provider reports them. |
| Duration | How long the run took. |
| Created At | When the run was started. |
Each row has three buttons:
- Copy ID copies the run ID, for example to look it up with
aiAgentRuns.showor to send to support. - Chat opens the conversation the run worked in. It is shown only for runs that belong to a session.
- Show opens the run's details.
Use the filters to narrow the list by AI Harnesses, Status and Created At (date range). The table does not update by itself; use the refresh button in its header to load new runs and status changes.
Run details
Show opens the AI Agent Run drawer.
Summary
| Row | Shows |
|---|---|
| ID | The run ID, with a copy button. |
| AI Harness | The harness name and type. |
| Status | How the run ended, or that it is still queued or running. |
| Error | The error message, for runs that did not complete. |
| Main Model, Fallback Model (after the soft limit), Fast Model (light background tasks) | The AI models the node set, for harnesses that use AI models. |
| Model (one of the harness's allowed models) | The model, for a hosted harness. |
| Created At, Started At, Finished At | When the run was queued, started and ended. |
| Duration | Time used / Max Duration (Seconds). |
| Model Calls | Calls used / maximum, with the soft limit when one is set. |
| Tool Calls | Calls used / maximum. |
| Tokens | Input and output tokens (and cache tokens), and total / Max Total Tokens. |
Steps
Steps lists what the agent did, in order, with the time since the run started on the right:
| Step | Shows |
|---|---|
| Model call n | The model, the input and output tokens, and either the tools the model asked for (Requested tools: …) or Final answer. |
| Tool call | The tool's name and how long it took. |
| Switched to the fallback model | The run reached the soft limit and continued on the fallback model. |
| Intermediate message | A message the agent wrote while working, as it was sent to the customer or added as a whisper. |
| Retry | A failed model call that was retried, with its error. |
If the harness reports no steps, the drawer says "No steps were recorded for this run."
Instructions, Input, Result and JSON Data
- Instructions: the node's instructions as the agent received them.
- Input: the task, with its tags already filled in.
- Result: the agent's final answer.
- JSON Data: the whole run record. Use the copy button to share it with your developer or with MonoChat support.
Debugging an agent
| What you see | What to check |
|---|---|
| Budget exceeded with Model Calls at the maximum | Look at the steps. If the agent keeps calling tools without converging, make the instructions and the tool descriptions more precise. If the task really needs more steps, raise Max Model Calls. |
| Budget exceeded with Tokens at the maximum | A long conversation history or large tool results fill the token budget. Lower Message History Limit, return less data from your tools, or raise Max Total Tokens. |
| Budget exceeded with Tool Calls at the maximum | The agent asked for more tools than allowed. Raise Max Tool Calls or tell it in the instructions which tool to use when. |
| Timed out | Check the tool call durations in Steps. A slow API or MCP server uses up Max Duration (Seconds). |
| Failed with "An AI scope or MCP tool was not found in this workspace." | A scope or MCP tool on the node was deleted. Update the node's Tool Settings. |
| Failed with "Choose one of the AI harness's allowed models." | The node's model was removed from the hosted harness's Allowed Models. |
| Failed with "This harness is not available on this server." | The harness needs MonoChat's agent runner, which is not available. Choose another harness. |
| The customer did not get the answer | Check that the run Completed, that Send answer is on in the node, and whether the Result is empty. |
| The customer saw "thinking" messages you did not want | Turn off Forward Intermediate Messages to the Customer; the messages then stay as whispers for your team. |
Compare a few runs of the same node before changing its budget. The steps show whether the agent needs more room or better instructions.