Skip to main content
Last updated on

OpenBox Dashboard

OpenBox dashboard at openbox.ai is where you configure policies and guardrails, and where you see the results of governance evaluation. This page explains what you will see in the dashboard after integrating the OpenClaw plugin.

Session Timelines

Each agent session appears as a workflow in the dashboard. The timeline shows the full lifecycle:

  • WorkflowStarted - when the session began
  • ActivityStarted / ActivityCompleted - each tool call and LLM inference, with timing and verdict
  • WorkflowCompleted / WorkflowFailed - how the session ended

Each activity entry includes:

  • Tool name or llm_inference
  • Input parameters sent to the tool
  • Output or error returned by the tool
  • Governance verdict (allow, block)
  • Duration
  • OTel spans (HTTP requests and filesystem operations that occurred during the activity)

Policy Configuration

Policies are configured in the OpenBox dashboard and take effect immediately - no plugin or agent restart needed.

Policies evaluate each tool call and LLM inference event and return a verdict. You define the rules; the plugin enforces the results.

Verdicts

VerdictEffect
allowAction proceeds normally
blockCurrent action is prevented; agent can continue with other actions

For details on how verdicts are enforced, see How It Works - Governance Verdicts.

Guardrails Configuration

Guardrails are also configured in the dashboard. They apply to LLM requests routed through the OpenBox gateway and evaluate input content for sensitive data or policy violations.

Available guardrail types:

TypeWhat it does
PII DetectionDetects and redacts personally identifiable information (emails, phone numbers, SSNs, etc.)
Content FilteringBlocks requests matching specific content categories
ToxicityBlocks abusive or hostile language from user interactions
Ban WordsCensors or blocks domain-specific banned terms (competitor names, internal codenames, regulated terms)

When you configure a guardrail:

  • It applies to all LLM requests passing through the gateway
  • No plugin configuration changes are needed
  • Results appear in the session timeline as part of the LLM activity events

Redaction behavior

When PII detection is configured in redaction mode, the dashboard shows:

  • The original content that triggered the guardrail
  • The redacted version that was sent to the LLM
  • Which guardrail rule matched

Usage Analytics

The dashboard provides visibility into:

  • Number of tool calls and LLM inferences per session
  • Governance verdicts over time
  • Guardrail triggers and redaction events
  • Session success/failure rates

What flows from the plugin to the dashboard

DataWhen it is sentPurpose
Session registrationFirst call in sessionCreates the workflow entry
Tool call + parametersBefore each tool executesPolicy evaluation
Tool result + OTel spansAfter each tool executesAudit trail, telemetry
LLM inference metadataOn each LLM input/outputUsage tracking
LLM request contentThrough gatewayGuardrails evaluation
Session outcomeOn session endWorkflow completion status