Skip to main content

​
Claude Observability

Airia integrates with Anthropic’s Compliance API and Claude OpenTelemetry (OTel) signals to give security teams a consolidated observability feed across Claude’s enterprise surfaces. With this integration, administrators can monitor Claude Code, Claude Cowork, and Claude.ai conversation content from Airia. Code and Cowork activity is routed through the Airia AI Gateway and inspected inline. Claude.ai conversation content—including chats, uploaded files, and projects—is captured through the Anthropic Compliance API and evaluated out of band. Airia correlates these signals into a single Claude Monitoring feed so security and compliance teams can review usage, evaluate activity against guardrails, and investigate policy violations from one place.

​
What’s Covered

Claude Observability brings together activity from:
  • Claude Code traffic routed through the Airia AI Gateway
  • Claude Cowork traffic routed through the Airia AI Gateway
  • Claude.ai conversations captured through the Anthropic Compliance API
  • Claude.ai uploaded files associated with monitored conversations
  • Claude.ai project metadata returned with Compliance API chat records
  • Claude OTel events, including prompts, model requests, responses, tool results, and errors
  • Airia guardrail evaluation outcomes for monitored Claude activity
Claude.ai conversation monitoring requires an enterprise Claude license and a Compliance API access key with the required Anthropic scopes.

​
How It Works

Airia uses two monitoring paths depending on the Claude surface.

​
Inline Inspection for Claude Code and Claude Cowork

Claude Code and Claude Cowork traffic is routed through the Airia AI Gateway. The gateway applies Airia guardrails inline and forwards Claude OTel events to Airia. These events are normalized into monitoring sessions, prompts, and raw events. Administrators can filter by user, service, surface, and date range to review usage patterns and investigate activity.

​
Out-of-Band Evaluation for Claude.ai Conversations

Claude.ai chats are collected through the Anthropic Compliance API. Airia syncs organizations, users, chats, messages, and file references from Anthropic, then stores the conversation content for monitoring and review. Because Claude.ai conversations do not flow through the Airia AI Gateway in real time, Airia evaluates this content out of band after it is synced. Guardrail results are attached to the same monitoring views used for OTel-backed activity, giving administrators a consistent review experience across all Claude surfaces.

​
Consolidated Claude Monitoring Feed

The Claude Monitoring feed combines Compliance API and OTel-backed data into one review surface. From the feed, administrators can:
  • View Claude activity by session, prompt, or raw event
  • Filter by date, user, service, and surface
  • Open a session to inspect the normalized conversation
  • Review guardrail outcomes attached to conversation turns
  • Inspect event details such as model, tool, token, and error signals when available
  • Access synced Claude.ai file content through Airia-controlled endpoints

​
Prerequisites

Before you start, make sure you have:
  • An Airia role with access to Claude Monitoring setup and monitoring feeds
  • An Airia API key for Claude Code and Claude Cowork OTel ingestion
  • An Anthropic Compliance API access key for Claude.ai monitoring
  • The required Anthropic Compliance API scopes:
    • read:compliance_org_data
    • read:compliance_user_data
  • Claude enterprise licensing for Compliance API access

​
Configure Claude Monitoring

  1. In Airia, open the Security area and go to Claude Monitoring.
  2. Select Configure.
  3. Choose the Claude surfaces you want Airia to monitor:
    • Claude Code
    • Claude Cowork
    • Claude.ai
  4. For Claude Code and Claude Cowork, select or create the Airia API key used to send OTel events to Airia.
  5. For Claude.ai, select the stored Anthropic credential that contains your Compliance API access key.
  6. Test the Claude.ai connection before saving. Airia verifies the credential without persisting test data.
  7. Select Save.
After setup, Airia starts ingesting configured OTel activity and syncing Claude.ai conversation content on the configured background schedule.

​
Review Claude Activity

Use Claude Monitoring to review activity across all enabled Claude surfaces.

​
Sessions

The session view groups activity into conversations or OTel sessions. Use this view to see who used Claude, which service produced the activity, how many prompts or events were recorded, and when the session was last active.

​
Prompts

The prompt view groups OTel-backed activity by prompt. Use this view to investigate tool calls, model requests, responses, errors, and other prompt-level signals.

​
Raw Events

The raw event view shows individual OTel events. Use this view when you need the most detailed record for troubleshooting, audit review, or policy investigation.

​
Conversation Details

Open a session to review the normalized conversation. Claude.ai sessions are built from Compliance API chat content. Claude Code and Claude Cowork sessions are built from OTel events. Guardrail outcomes are attached consistently in the conversation view regardless of source.

​
Security and Privacy

  • Anthropic Compliance API credentials are stored in Airia’s credential system and resolved server-side.
  • Claude.ai uploaded file content is fetched through Airia-controlled endpoints. The browser does not receive the Compliance API access key.
  • File content responses use defensive response headers to prevent browser MIME sniffing and direct inline rendering of uploaded content.
  • Compliance sync is incremental and uses retry-aware behavior for rate limits and transient upstream failures.
  • Guardrail evaluation is tenant-scoped and controlled by Airia feature flags for Claude surfaces.

​
Troubleshooting

Claude Code or Claude Cowork activity does not appear
  • Confirm the selected Airia API key is active.
  • Confirm the local Claude OTel configuration points to the Airia ingest endpoint.
  • Check that the service selected in setup matches the surface you are testing.
Claude.ai conversations do not appear
  • Confirm the tenant has an enterprise Claude license with Compliance API access.
  • Confirm the stored Anthropic credential is an Anthropic credential type and contains a valid API key.
  • Confirm the key has the required Compliance API scopes: read:compliance_org_data and read:compliance_user_data.
  • Use Test connection in Claude Monitoring setup to verify the credential.
  • Allow time for the background sync cycle to collect new or recently updated chats.
Need help? Contact Airia support.