Hyperparam

A SaaS platform that captures AI model calls, prompts, tool calls and session traces and analyzes them at scale to answer where token spend goes, which prompts and tools work, and why sessions fail. It is for engineering teams and data teams at B2B organizations that run AI agents, coding assistants and chatbots. It is positioned against observability dashboards and warehouse SQL that aggregate or cannot read nested conversation payloads, by storing data as open Iceberg tables in storage the customer owns and querying directly in the browser.

Key features

• Capture every model call via collector
• Store traces as Apache Iceberg tables
• Read millions of rows from S3/GCS/Azure
• Search logs by keyword, meaning and SQL
• Run language model over matching rows
• Browser-based querying with no cluster
• Filter, search and cluster traces
• Generate derived columns at scale
• Build SQL views across sources
• Save reusable analyses as skills
• Classify tool failures and suggest fixes
• Score and filter low-quality responses
• Categorize prompts and user messages
• Compare model outputs side by side