Running an AI Agent platform is not only about whether the model answers. Teams also need to know where token cost is going, whether queues and responses are healthy, what users are doing in sessions, and whether risky operations are appearing in the workflow.
The source article introduces OpenClaw Usage Insights, a Tencent Cloud CLS-based monitoring package for OpenClaw. Its purpose is to connect OpenClaw runtime data into CLS and expose prebuilt dashboards for cost governance, operations monitoring, session management, session detail analysis, security audit, and raw log search.
| Operator question | CLS capability | Practical value |
|---|---|---|
| Where is money being spent? | Cost governance dashboard | Summarizes total cost, total token usage, average session cost, and cost distribution so expensive sessions can be located quickly. |
| Is the system healthy? | Operations monitoring dashboard | Tracks message processing, queue depth, execution latency P95, card distribution, log series, and OTEL-style runtime metrics. |
| Where are users active? | Session management dashboard | Summarizes session volume, average turns, tool-call count, channel distribution, and model distribution. |
| What happened inside one session? | Session detail dashboard | Reconstructs the full interaction flow with token consumption, per-turn details, single-message cost, problem checks, and prompt optimization clues. |
| Is there a security risk? | Security audit dashboard | Counts high-risk sessions, traces high-risk command execution, and audits sensitive-file access. |
| How do I inspect raw logs? | Log search | Filters server-side logs by instance, conditions, or AI-generated query statements. |
Prerequisites from the source article
The onboarding path is intentionally simple, but the source still lists three requirements:
- OpenClaw has already been installed and is running.
- Tencent Cloud CLS has been activated.
- A Tencent Cloud API key is available, including
SecretIdandSecretKey.
The article says the CLS Application Center can connect Tencent Cloud CVM and Lighthouse servers through a visual flow. Self-managed machines outside Tencent Cloud can be connected through a one-command deployment path.
Step 1: open the OpenClaw app in CLS
The first screenshot shows the CLS console with OpenClaw selected under the Application Center. The operator opens the OpenClaw integration page, where the main action is to start access management and connect OpenClaw hosts.
The list view distinguishes machines that have already been collected from machines that have not. This matters because the next step should target only uncollected hosts, avoiding duplicate collector deployment.
Step 2A: connect Tencent Cloud CVM or Lighthouse instances
For Tencent Cloud-hosted machines, it have four concrete actions:
- choose the non-self-managed server type, either
Cloud Server CVMorLighthouse; - select instances whose OpenClaw collection status is
not collected; - enter
SecretIdandSecretKey; - click
Next.
The article says the system then installs the required collector automatically. Operators do not need to manually log in to each server for this path.
The completion screen means the automated installation has finished. After that, the host should appear in the OpenClaw access-management list as connected.
Step 2B: connect self-managed servers
For self-managed servers, the console generates a command after the operator selects the region and provides the Tencent Cloud API credentials. The important thing is: copy the generated command and run it on the target server where OpenClaw is installed.
Instead of a GUI-only install, the server runs a generated command and returns installation progress in the shell. This is useful when OpenClaw is deployed outside Tencent Cloud infrastructure.
After connection, the access-management page becomes the inventory view. It shows which OpenClaw machines are already connected and available for dashboards and log search.
Step 3: open prebuilt dashboards
After onboarding, go to CLS Console -> Application Center -> OpenClaw -> Dashboard, then select the server instance. CLS provides five dashboard categories out of the box.
Cost governance dashboard
The cost dashboard turns token consumption into operational metrics. The visible cards include total cost, total token usage, average session cost, and cost distribution. The charts below break cost down by dimension so teams can move from "our token bill is high" to "this session, model, or usage pattern is driving the spend."
Operations monitoring dashboard
The operations dashboard is aimed at 24/7 runtime visibility. In the source article, it is used to detect queue backlog, degraded response time, and sudden error growth. In English terms, this is the "is OpenClaw healthy right now?" dashboard.
Session management dashboard
The session dashboard summarizes user interaction behavior. It covers session count, average turns, tool invocation frequency, channel distribution, and model distribution. This is useful for product and operations teams that need to understand how users are actually interacting with the agent system.
Session detail dashboard
The session detail view is a troubleshooting layer. It reconstructs a single conversation, including token usage, per-round interaction details, and problem checks. The source article positions this dashboard as useful for issue diagnosis, prompt optimization, and session analysis.
The source shows two entry paths. Operators can click a session ID or session content row from the session-management overview, or open the session-detail dashboard directly and filter by server instance and session ID.
Security audit dashboard
The security audit view focuses on risky sessions, high-risk command execution, and sensitive-file access. For an AI Agent operations team, this is the part of the dashboard set that turns raw agent behavior into security review signals.
Search raw OpenClaw logs
Beyond dashboards, operators can open Log Search inside the OpenClaw application page. The search page supports selecting an instance, adding filter conditions, or using AI to help write a query statement.
The result view keeps raw logs and statistical charts together. This makes the flow practical: start from a dashboard anomaly, jump to log search, filter by the relevant server or session condition, and inspect the original event records.
What this gives an OpenClaw operations team
OpenClaw Usage Insights is best understood as an observability package for AI Agent operations:
- cost governance for token spend and expensive sessions;
- runtime monitoring for queues, latency, errors, and system health;
- session analytics for user behavior and model/tool usage;
- single-session tracing for debugging and prompt optimization;
- security audit for risky commands and sensitive actions;
- raw log search for evidence-level troubleshooting.
The source article also previews two future directions: an AI-powered analysis center for automatic diagnosis and optimization, and a broader intelligent alerting center for cost, performance, and security thresholds.
Top comments (0)