Skip to main content
Darwin is in development and not yet generally available. This page previews what it will do.
Darwin is Vijil’s evolution engine. Where Diamond tests an Agent before deployment and Dome protects it in production, Darwin closes the loop by learning from what happens in the real world and feeding those lessons back into the Agent. Darwin uses reinforcement learning over production telemetry, the real interactions, failures, and edge cases your Agent encounters, to continuously improve its resilience as behavior drifts over time.

​
How Darwin Fits In

Diamond, Dome, and Darwin form a continuous cycle over an Agent that Discover can find:
  • Diamond evaluates the Agent before deployment and produces a Trust Score.
  • Dome protects the Agent at runtime and logs what it sees.
  • Darwin learns from that production telemetry and proposes improvements, which feed back into a stronger Agent.

​
The Adapted Stage

An Agent reaches the Adapted stage once at least one Darwin proposal has been accepted following an evolution run. This marks an Agent that has not only been tested and protected, but improved based on evidence from real-world use.
Last modified on October 2, 2026