rpk ai trace list
List traces of agent runs and AI gateway calls, newest first unless
--order-by says otherwise.
The filter flags build ListTraces' AIP-160 filter: values within one flag are
ORed, different flags are ANDed, and --filter is ANDed with all of them. The
server allows one constraint per field, so --filter must not repeat a field a
flag already sets. --since is required and the window is at most 30 days,
so no listing reads all of history.
--limit caps the traces printed (default 50, at most 1000), read in pages of
up to 100. -o json and -o yaml print the ListTracesResponse.
| This command was introduced in ai version 0.2.94. |
Examples
This section provides examples of how to use rpk ai trace list.
Failed runs and runs with a failed tool call in the last day
rpk ai trace list --since 24h --errors
One agent’s running traces
rpk ai trace list --since 24h --agent support --status running
Most expensive traces this week
rpk ai trace list --since 7d --order-by "usage.cost.total_microcents desc" --limit 10
Flags
| Value | Type | Description |
|---|---|---|
|
string |
agent name or uid. (New in 0.2.100) |
|
string |
conversation id. (New in 0.2.100) |
|
bool |
failed runs and runs with a failed tool call. (New in 0.2.100) |
|
string |
raw AIP-160 filter, ANDed with the flags above. |
|
int |
maximum traces to print, 1 to 1000. (New in 0.2.100) |
|
string |
AIP-132 order_by, for example, |
|
string |
traces started at or after, as RFC 3339 or a duration ago (15m, 24h, 7d); required and at most 30d before |
|
strings |
running, completed, error or incomplete ( |
|
string |
traces started before: RFC 3339 or a duration ago. (New in 0.2.100) |
|
strings |
caller email or end-user id ( |
Global flags
| Value | Type | Description |
|---|---|---|
|
string |
Redpanda or |
|
stringArray |
Override |
|
bool |
Ignore |
|
string |
|
|
bool |
Enable verbose logging. |