LightOn

Search as part of an AI system

LightOn builds enterprise AI search software. Its platform connects organizational documents and other knowledge to search interfaces and agent workflows. This places it in the retrieval layer of AI applications: finding relevant material before an assistant or agent uses it. The current platform is presented through LightOn Core and Apps. [1]

Its research also contributes model artifacts. LightOn co-developed ModernBERT with Answer.AI in December 2024. The release supplied encoder models intended for retrieval, classification and other language-understanding tasks. LightOn's announcement explains the work in terms of searching enterprise knowledge before passing selected material to a generative model. [2]

Encoders have a different job from chat models

ModernBERT-base is an Apache 2.0 masked-language-model checkpoint. It can be adapted for tasks such as classification and retrieval, rather than generating conversational replies as a decoder model would. That makes LightOn's contribution relevant to the components surrounding a language-model application as well as the application itself. [3]