Watch Demo Video —>

Athina Monitor assists developers in several key areas:

  • Visibility: By logging prompt-response pairs using our SDK, you get complete visibility into your LLM touchpoints, allowing you to trace through and debug your retrievals and generations.
  • Usage Analytics: Athina will keep track of usage metrics like response time, cost, token usage, feedback, and more, regardless of which LLM you are using.
  • Query Topic Classification: Automatically classify user queries into topics to get detailed insights into popular subjects and AI performance per topic.
  • Granular Segmentation: You can segment your usage and performance metrics based on different metadata properties such as customer_id, prompt_slug, language_model_id, topic, and more to slice and dice your metrics.
  • Data Exports: Export your inferences to CSV or JSON for external analysis.

Here are some examples of how Athina can help you understand your LLM behavior:

  • For queries related to Refunds, retrieval accuracy is only 55%. * 80% of user feedback is negative for customer ID nike-california-244 * Your model’s answer relevance is only 37% for prompt customer_support/v2.5 * Your avg. response time is 6.7s for gpt-4, and 2.1s for gpt-3.5-turbo * You are spending an average of $381.44 per day on OpenAI inferences