“I think that photos are an underappreciated source of information,” Mayer said. “You’ll be surprised what we can learn about you and how good a job we can do with your photos.”
The product is built around two main experiences. Get It Done focuses on recent photos and screenshots, looking for tasks Dazzle can help complete. A photo of a jacket, for example, can be used to identify the item, locate where it is sold and help purchase it. Images of a flat tire, damaged flooring or a broken garage door can prompt Dazzle to find a suitable professional and arrange a repair.
The assistant can also extract information from school flyers, invitations and appointment cards and add dates to a calendar. QR codes can be saved to a wallet.
Ideas takes a broader view of a user’s photo history. Dazzle analyzes previous images and screenshots to generate personalized suggestions around restaurants, concerts, gifts, activities and travel. The company says users can then act on those suggestions with only a few additional steps.
Mayer said Dazzle can infer details ranging from hobbies to family interests based on the images people already keep on their phones.
“We understand whether or not you like to ski, where your most recent trip was, what types of things your kids are into,” she said.
Dazzle’s chat experience uses frontier AI models combined with the personal context extracted from photos. The assistant can use that information to answer questions such as what gift might suit a family member or how someone could spend a weekend.
The product is available through an iOS app, the web, iMessage and text messaging. The web version is intended for larger tasks such as vacation planning, where Dazzle can suggest destinations, accommodations, activities and an itinerary.
Dazzle also offers an MCP connection that can bring its photo-based context into Claude, Codex and OpenClaw.
The company is emphasizing privacy as it asks users to grant an AI access to potentially large photo libraries. Photo access is optional, and users can choose between sharing an entire library, selected images or nothing at all.
Dazzle says images it determines may contain sensitive information are excluded immediately. Personal data is used to personalize the product for the individual user, according to the company, and is not used to train public AI models, sold or rented, or used for third-party advertising targeting.
The company also says its agreements with outside AI providers prevent those companies from using Dazzle users’ personal information to train their own models.
Dazzle has taken a similar approach to account access. The company describes its agents as acting “on your behalf, not as you,” with their own accounts and tools rather than broad access to a user’s credentials.
Mayer founded Dazzle AI in 2025 after previously leading Yahoo and working at Google. The company describes its broader goal as reducing the gap between what people want to accomplish and what AI can help them do.
Dazzle raised an $8 million seed round in December 2025 at a $35 million post-money valuation. Forerunner’s Kirsten Green led the financing, with Kleiner Perkins, Greycroft, Offline Ventures, Slow Ventures and Bling Capital also participating.
At the time, Mayer said increasingly reliable foundation models were shifting more opportunity toward consumer applications.
“Now that foundational models have reached a level of consistent excellence, they’ve become a reliable infrastructure. The new frontier is applications—leveraging that power to create real, tangible value.”
Dazzle is initially available by invitation. Users can request access through its iOS app, its website or by texting the company’s signup number.
The launch puts Dazzle among a growing group of consumer AI assistants, but its defining bet is narrower: instead of learning primarily from email, calendars or other text-heavy accounts, it treats the camera roll as the starting point for understanding the user.
This analysis is based on reporting from unite.ai.
Image courtesy of Dazzle.
This article was generated with AI assistance and reviewed for accuracy and quality.
