Streaming Token Counting

The connect agent parses Server-Sent Events (SSE) streaming responses to capture accurate token usage.

• Extracts actual token counts from the usage field in streaming responses from OpenAI and Anthropic APIs.
• When the API response does not include usage data (e.g., some streaming modes), the agent falls back to content-based estimation using token approximation.
• Token counts are included in the audit event sent to AgentWatch, ensuring accurate cost tracking for streaming requests.