Summary: Sesame, a voice AI startup put together by Oculus’ founders and several early team members, placed an iOS preview on the App Store on May 27, 2026, opening in 39 countries and free at first. ChatGPT-style products take the shape of an “input box plus conversation history.” Sesame has been built around voice from its first day: the whole experience is organized by how speech paces itself and when to speak; on-screen visuals only support.
Sesame, a voice AI startup put together by Oculus’ founders and several early team members, placed an iOS preview on the App Store on May 27, 2026, opening in 39 countries and free at first. The date lands one year to the day after the company’s web Research Preview — the preview that pulled in more than a million people to talk with voice agents Maya and Miles across its opening weeks, by investor Sequoia’s count. Simone and Charlie join the returning pair, completing a four-agent roster in which every member carries a distinct voice, personality, point of view, and memory.
ChatGPT-style products take the shape of an “input box plus conversation history.” Sesame has been built around voice from its first day: the whole experience is organized by how speech paces itself and when to speak; on-screen visuals only support. Even the terminology is a statement — the company refuses the chatbot label and keeps agent.
Four Agents, Four Separate Memories
Character-first is the wager Sesame has placed. Maya, Miles, Simone, and Charlie each received a clearly defined voice, personality, and point of view, and their memories never cross — whatever you said to Miles remains with Miles alone. Memory does travel between voice and text, so typing requires no fresh run-through of context. Privacy gets an Incognito Mode: agents can still read earlier context while storing nothing new, and it all disappears once the conversation ends. The setup deliberately mirrors real social life — different friends hear about different topics, and picking another agent means picking another context rather than pouring it all into a single faceless assistant. Whether memory built this way actually retains users is the piece of the wager still awaiting proof.
Talking and Searching at the Same Time: Parallel Retrieval With Mid-Sentence Course Corrections
Latency is where the engineering gets genuinely interesting. Sesame assembled quick search and retrieval machinery: an agent fires off several parallel searches while still talking, folding whatever it finds into the answer as it goes — even reversing course mid-sentence. In the company blog’s telling, this is what lets agents run “slower and smarter agentic loops without awkward interruptions.” TechCrunch relays the team’s framing of the speed-versus-quality tradeoff: “There’s an inherent tension between replying quickly and taking the time to compose thoughtful responses,” and “a slower response is usually more correct, but it can also feel unnatural if it takes too long.”
The interface keeps voice first and visuals supporting: search returns as cards with images, takeaways go into Notes, Deep Dives carry the in-depth results, and a texting mode covers moments when speaking aloud won’t work. As architecture, this tucks thinking time inside the flow of speech — the agent begins talking instead of leaving dead air while retrieval finishes, then adjusts course as facts arrive. If it holds up, it rewrites the latency users perceive and the retrieval bill of every conversation.
Free for Now, 39 Countries Wide, Starting From a Million Users
On the business side, the app costs nothing “for the time being,” sign-up may pass through a brief waitlist, and an Android preview is also in the works. The funding runs deep: in October 2025, just as the beta opened, Sesame announced a $250 million Series B with Sequoia leading it; Sequoia adds that the Research Preview crossed a million users in its opening weeks. By consumer voice-AI standards that is a strong start, and Sesame’s own yardstick is openly the traditional text-chatbot experience ChatGPT made mainstream.
Going live in 39 countries in one stroke carries meaning too. Text tolerates accent variation and localization gaps far better than voice does, so a first wave stretched this wide reads as genuine confidence in how the system handles many languages — or at least a readiness to put the wager on voice early.
From “Think With You” to “Do For You” — and 2027 Eyewear
In Sesame’s telling, the iOS app is step one on the road to intelligent eyewear slated for 2027. For people who made their careers in VR, the route needs little justification: once the device is a pair of glasses, voice is the only input that feels natural. The company has also flagged where things go next — agents graduating from “think with you” to “do for you,” hence the insistence on the word agent.
Product teams should keep two signals in view. First, does the persona-driven, memory-per-agent design genuinely keep people around, or does it fade with the novelty? Second, will “keep searching while speaking” harden into the standard architecture for voice products, resetting the latency users tolerate and what every session costs? The money question remains open — subscriptions, enterprise licensing, or hardware bundling will eventually have to reach the table, and the 2027 eyewear deadline will show who in this voice-agent wave is serious.