Basalt

Basalt is an analytics and learning infrastructure layer designed specifically for AI agents and AI-powered products. It captures detailed customer behavior around how users interact with your agents, then feeds those insights back into your systems so the agents can improve over time. The primary purpose of Basalt is to provide a robust feedback and experimentation framework that makes AI agents more reliable, effective, and aligned with real user needs.

Basalt typically offers event-level tracking of conversations and user actions, session replay or interaction histories, and structured logging of model inputs and outputs. It supports labeling and annotation workflows so teams can mark successful, failed, or problematic interactions for downstream training or fine-tuning. The platform often includes evaluation tools for comparing different prompts, models, or policies, enabling systematic A/B testing and regression detection. By centralizing this data and feedback, Basalt helps teams close the loop between deployment, observation, and continuous improvement of their AI agents.