Skip to content

Runs

Early Access

Spyglass is currently in early access with trusted partners. Contact us for early access.

Spyglass Runs#

A Run in Spyglass is the process of evaluating an AI system by sending either a series of prompts or Datasets to specific Targets and measuring the results with Scorers. Runs are a core component that enable you to:

  • Simulate potential attacks and manipulations
  • Test model resilience under adversarial conditions
  • Generate actionable insights about vulnerabilities
  • Compare performance across multiple models
  • Download results for further analysis

Each Run consists of a Dataset, Target, and Scorer combination within a Project. The results help identify security risks and vulnerabilities in your AI applications during both development and deployment.

Following a run execution, datasets are available to download which includes inputs from the original dataset and model responses in a new column. Additionally, you can navigate through the run logs within the Spyglass UI by navigating to Spyglass > Projects or Spyglass > Runs in the navigation bar:

spyglass runs count spyglass run summary

spyglass magic commands run samples