This topic describes how to use Agent observation and its limitations. Agent observation is an eBPF-based observability tool for AI agents. It collects and correlates fine-grained data across the entire agent runtime without any intrusion into your business logic. It primarily collects token consumption data for each session.
Limitations
-
Regions
This feature is available only in the Chinese mainland and Hong Kong (China) regions.
-
Data storage
Data stored in SLS can be queried for up to 7 days. Data older than 7 days is automatically deleted.
-
Operating systems
Architecture
Operating system
x86
-
Alibaba Cloud Linux 4
-
Alibaba Cloud Linux 3
-
Alibaba Cloud Linux 2
-
Ubuntu 22.04
-
Ubuntu 24.04
-
-
Instances
This feature applies only to instances that run Openclaw or Hermes applications.
-
Other limitations
Token statistics for Coding Plan are not supported. If you use Coding Plan, the input and output token counts may be zero.
Enable Agent observation
Step 1: Create a configuration
-
Go to the Operating System console.
-
In the left-side navigation pane, choose Component Management.
-
Create a configuration and give it a name related to Agent observation.
Step 2: Activate the configuration
-
In the left-side navigation pane of the Operating System console, choose System Management.
-
In the managed plan for the target instances, apply the configuration from the previous step to the SysOM component.
-
Click Submit. Agent observation is now enabled.
Token analysis
You can find this feature on the Agent observation > Token analysis page in the Operating System console.
-
Go to the Operating System console.
-
In the left-side navigation pane, choose Agent observation > Token analysis.
-
Configure the query parameters.
Parameter
Description
Time range
Select the time range to view token usage.
Instance ID
Select the instance where the AI agent application is deployed.
Data source
Select the process of the AI agent application.
-
Click Query. The token usage results panel appears.
The results panel contains the following information:
-
Sessions: Displays the number of sessions, the total number of input tokens, and the total number of output tokens.
-
Token time-series chart: Displays token usage trends by input, output, and total, as well as a token time-series chart by model.
-
Session records: Displays token usage records by session.