November 2025 Summaries

Perplexity has introduced enhanced personalization and memory features for its Comet Assistant, designed to retain users’ preferences, interests, and relevant conversation details across interactions. Rather than relying solely on broad language-pattern predictions, the system retrieves pertinent information from a user’s encrypted memory store to produce more precise and individualized responses, such as travel suggestions, shopping recommendations, or follow-up advice. The feature aims to reduce the need for users to repeatedly provide context or manually summarize past discussions, while allowing context to persist across the different AI models available on the platform. Users retain control over memory settings, including the ability to disable memory, use incognito mode without saving memory or search history, and opt out of data use for model improvement.

Perplexity has launched a free AI-powered shopping experience for U.S. users that aims to improve online product discovery through conversational, personalized search. The service uses users’ search history and stated preferences to maintain context across follow-up questions, recommend products suited to individual needs and tastes, and present focused product cards with relevant specifications and reviews rather than generic product grids or affiliate-driven lists. Through a partnership with PayPal, users can complete purchases directly within the Perplexity experience from participating retailers, while merchants remain the merchant of record and retain customer, return, loyalty, and post-purchase relationships. The feature is available on desktop and the web, with iOS and Android availability planned in the coming weeks.

Perplexity has launched Comet for Android, extending its AI-native browser from desktop to mobile with a design intended specifically for mobile use rather than a scaled-down desktop experience. The browser integrates the Comet Assistant, which can help users research, summarize, shop, learn, and perform assigned tasks while displaying its actions and allowing user intervention. Key Android features include voice-based interaction with the assistant across open tabs, cross-tab content summarization, contextual answers directly on webpages, tab memory, and a built-in ad blocker with site whitelisting controls. Perplexity presents Comet as a private, distraction-reducing browsing tool designed to support user curiosity and productivity by making AI assistance available throughout the web.

Perplexity announced a government-wide partnership with the U.S. General Services Administration to make its Enterprise Pro for Government AI platform available to federal agencies and servicemembers through the GSA Multiple Award Schedule. Agencies can obtain the service at essentially no cost for up to 18 months, followed by negotiated long-term pricing, with the company describing the agreement as the first direct government-wide contract secured by a major AI company. The platform provides access to multiple AI model families, including Perplexity Sonar, GPT, Claude, and Gemini, alongside configurable access to Perplexity’s knowledge index and agency systems, allowing organizations to select models and information sources based on mission needs. Perplexity also emphasizes its security posture, citing secure-by-default protections for federal users, an AI Prioritization designation from GSA, and an expedited path toward FedRAMP authorization through the cloud-native 20x process.

Perplexity has announced upgrades to its Comet Assistant, emphasizing an approach to agentic AI centered on transparency, user control, sound judgment, and accurate information. The assistant now makes its browsing activity visible through a sidecar that shows actions such as clicking, scrolling, and interacting with websites, while providing options to stop or redirect it. Users can decide whether to browse independently, allow Comet to act for a single task, or enable automatic assistance when it detects an opportunity to help. For sensitive tasks such as logging in or making purchases, Comet is designed to pause and request permission before proceeding. Perplexity presents these safeguards as principles shared across its other AI products, including Deep Research, Labs, Email Assistant, and Background Assistants, and describes Comet as a private, personalized tool intended to earn users’ trust as its capabilities expand.

Comet has launched a redesigned version of its AI Assistant for all users, expanding its ability to handle longer, more complex web-based tasks across multiple browser tabs. The company says the assistant has improved tools for interpreting and interacting with websites, can perform a wider range of actions, and is particularly useful for workflows involving information from several sources, such as transferring data between websites and spreadsheets. Suggested uses include organizing job searches, comparing travel prices, and compiling school attendance records into Google Sheets. Comet reports that internal testing found the updated assistant performed 23% better than the prior version on complex tasks, while new permission controls allow users to decide whether it may interact directly with their browser and apply that choice throughout a task.

TransferEngine is a Rust-based, open-source RDMA communication library developed by Perplexity AI to support flexible point-to-point communication for emerging LLM workloads such as disaggregated inference, Mixture-of-Experts routing, and asynchronous reinforcement-learning post-training, where conventional collective communication frameworks can be restrictive. Designed to operate across NVIDIA ConnectX-7 and AWS Elastic Fabric Adapter hardware, it assumes reliable but potentially unordered transport, uses an ImmCounter mechanism for order-independent completion notifications, and transparently distributes transfers across multiple NICs. Its compact API supports registered memory, two-sided control messaging, zero-copy contiguous and paged writes, scatter and barrier operations for peer groups, and UVM Watchers that coordinate GPU computation with CPU-initiated RDMA transfers. Perplexity reports using the library to independently scale prefill and decode services through layer-by-layer KV-cache transfers, synchronize model weights for RL post-training in under two seconds, and enable low-latency trillion-parameter MoE operations on AWS EFA.

Perplexity argues that Amazon’s legal threat seeking to prevent Comet AI assistant users from shopping on Amazon represents an effort to restrict consumer choice and emerging agentic AI technology. It distinguishes user agents from crawlers or bots, describing them as private, user-directed assistants that operate with a person’s own credentials and permissions to compare products or make purchases. Perplexity contends that Amazon’s opposition is motivated by protecting advertising, sponsored listings, and upselling opportunities, despite Amazon’s stated interest in partnerships with third-party AI agents. The company frames the dispute as part of a broader question over whether AI will empower individuals to control their online experiences or enable large corporations to retain influence over consumer decisions, and says it will continue defending users’ ability to employ capable, personal AI assistants.