Firefox 154.0 Ships Zero-Retention AI Search via Exa

Mozilla launched Firefox 154.0 on August 18, 2026, introducing a major update to its Smart Window feature set. According to Mozilla’s announcement and rollout details, the release centers on a new partnership with Exa (formerly Metaphor) to power real-time search, alongside automated tab management and visual history retrieval. The rollout targets beta testers on Windows and macOS, with deployment to Linux still resolving initial release inconsistencies.

Exa Integration and Tab Management

The Exa partnership shifts Smart Window from a static chat interface to an active retrieval engine. Users can pull real-time web data and source citations directly within the chat interface, eliminating the need to context-switch to a separate search engine results page. For mobile users, this capability extends to Quick Answers on Firefox for iOS across both iPhone and iPad.

The update also tackles browser state management through automatic tab grouping. Firefox analyzes the text and context of open pages to suggest categorized groups and automatically proposes names for these clusters. The system includes an automatic duplicate detection feature that allows users to close redundant tabs with a single click.

Visual History Retrieval

Instead of querying exact URLs, users can now search their past browsing sessions using natural language. A query for specific items, such as outdoor equipment viewed days prior, surfaces visual snapshot previews rather than a text list of domains. This shifts history navigation from a string-matching exercise to semantic retrieval. This implementation mirrors broader industry shifts where developers evaluate approaches for adding long-term memory to agents using multimodal context rather than pure text ingestion.

Zero-Retention Architecture

Mozilla distinguishes Smart Window through strict data boundary enforcement, positioning the tool as a direct alternative to integrations like Microsoft Edge’s Copilot and Google Chrome’s Gemini. The implementation relies on a zero-data-retention policy for all chat interactions. Neither Mozilla nor Exa uses session context or chat history to train their underlying models.

The feature remains strictly opt-in, functioning as a distinct mode separate from classic browsing windows. Users must explicitly toggle the AI features via the top-right interface icon, the File menu, or a newly added “Manage AI” action located in the address bar.

Local State Storage

To enable semantic history search without compromising its privacy stance, Firefox stores session Memories locally on the device. This architecture avoids transmitting comprehensive browsing logs to cloud inference endpoints. For engineering teams evaluating local privacy audits, this physical data separation provides a verifiable mechanism for keeping user context off centralized servers.

Early evaluations from technical reviewers highlight the utility of this localized approach. Testers report the tool executes tab summarization and comparison without the heavy resource bloat typical of competing AI sidebars, though the effectiveness heavily depends on the depth of local browsing context available to the system.

If you manage complex research workflows across dozens of tabs, Smart Window provides a structured way to consolidate state without exposing your browsing history to third-party model training. You should review the “Manage AI” settings to ensure your local storage parameters align with your hardware constraints.