Mozilla.ai is making the search layer behind AI agents configurable in Otari, giving developers a choice between Mozilla’s service, their own search-provider key, or a self-hosted backend. The change addresses a quiet privacy and procurement issue: an agent’s searches can contain internal service names, customer research, and details about unfinished products.
Many popular agent tools expose a generic web-search function without naming a configurable search provider. Mozilla notes that this opacity also appears in some local-model tools, where model inference may run locally but web queries still travel to a hosted endpoint. That means teams may not know which contracts, retention rules, or per-query prices govern their data.
Otari’s approach separates the agent and model from the search provider. Teams can keep their existing model while selecting where queries go and how the search service is operated. This does not make web research private by itself: the chosen provider still receives requests, and self-hosting adds operational work. It does make that dependency visible and replaceable instead of leaving it as an undocumented default.