Runs
| Overview | |
|---|---|
| What it is | The record of each agent execution: model calls, tool calls and result |
| CLI | agt run, agt agent run |
| Web console | Execution bar, the record under each reply, Debug (调试模式) |
| Settings | scheduler.model_steps |
| Storage | runtime/runtime.db |
| Crates | agt-runtime, agt-inference |
What it does
Every time an agent replies to a message or works on a task, agt records a run: its input, status, model calls, tool calls and output. Runs let you see what an agent did, why it failed, and what it cost in calls.
A native run loops between the model and tools until the model gives a final answer. Messages that arrive meanwhile are delivered at the next tool boundary (steering) or after the current work (follow_up). Codex runs are a Codex turn, with Codex's own tool loop.
completed means the model finished normally, not that the goal was achieved; check the tool failures and the output. A run cut off during a model or tool call is never retried automatically, because the call may already have had an effect.
How to use it
Web console: each reply shows its tool calls, duration and failures; click to expand or open the full record. Turn on Debug for run IDs, call counts and error codes.
CLI:
agt run list --limit 20
agt run info <run>
agt run calls <run>
agt run tool-detail <run> --activity <activity>Common parameters
| Parameter | Meaning |
|---|---|
agt run info <run> | Status, input, output, errors, tool failure count |
agt run calls <run> | Each model and tool call, with timing and status |
All parameters
agt run commands
| Command | Options | Meaning |
|---|---|---|
list | --limit <n> (default 50), --conversation <id> | Recent runs; with --conversation, only top-level runs triggered there |
info <run> | One run | |
calls <run> | --after <id> (default 0), --limit <n> (default 50) | Model and tool calls. Records the tool name, timing, status, safe error code and the shape of the arguments, not their values. Page with the last ID as --after. |
children <run> | Runs delegated from this run (agt agent run --delegate) | |
tool-detail <run> | --activity <id> (required) | Full arguments and result of one tool call, truncated at 64 KiB per field. Local operator only. |
agt agent run
Runs an agent once, outside any conversation. Options are listed in Agent.
Settings
| Key | Default | Meaning |
|---|---|---|
scheduler.model_steps | null | Maximum model calls in one native run. null for unlimited. Does not limit Codex. Takes effect for new runs. |
Run status
| Status | Meaning |
|---|---|
running | In progress, or cut off before a final status was recorded |
suspended | Waiting with a saved checkpoint (tasks only) |
completed | The model finished |
failed | An error ended the run |
canceled | Stopped by a person, a disabled agent or an ended discussion |
interrupted | Stopped by a crash or restart, and the stop is confirmed. Check its effects before trying again. |
outcome_unknown | Cut off during a step whose effect cannot be confirmed; check before retrying |
Call record status
| Status | Meaning |
|---|---|
started | The call began and no result was recorded: the outcome is unknown |
succeeded, failed, canceled | The call's result |