Hodoscope
What it is
Hodoscope is a Python library and CLI tool for unsupervised analysis of AI agent trajectories. It extracts actions from evaluation logs, summarizes them using LLMs, embeds the summaries into vector space, and visualizes them as interactive 2D plots to identify behavioral patterns and differences between models.
The problem it solves
Reading raw agent trajectory logs one-by-one does not scale when evaluating multiple models or configurations, making it difficult to identify patterns or anomalies in agent behavior.
What you could use it for
- Comparing behavioral differences between two LLM models on the same evaluation task
- Visualizing clusters of agent actions to identify common failure modes or strategies
- Analyzing large volumes of agent logs from Inspect AI, OpenHands, or Docent formats
Read the source ↗Back to the library →
Summary written by an AI model from the source; check the source before you rely on it.