Sesame Launches Four Eerily Realistic AI Voice Agents Ahead of Smart Glasses
Sesame opens its four voice agents to everyone on iOS and Android, with smarter models, app connectors, and glasses teased for 2027.
- Sesame launched its voice assistant apps on iOS and Android with four distinct AI characters.
- Maya, Miles, Simone, and Charlie each have their own voice, personality, and persistent memory.
- Agents can now browse the web, run tasks, and connect to Gmail, Calendar, and Drive.
- New skills feature lets users build plain-language playbooks that run on demand or on schedule.
- App is a stepping stone to Sesame's Made-in-Japan smart glasses, planned for 2027.
- Free during rollout; English is the only officially supported language today.
Sesame launches four AI voice agents ahead of its planned glasses
Sesame has moved Maya, Miles, Simone, and Charlie from preview into general availability on mobile. The release adds web research, connected apps, scheduled workflows, persistent memory, and a software foundation for the company’s planned AI glasses.
A preview becomes a product
Sesame is a San Francisco startup co-founded by former Oculus CEO Brendan Iribe. Its early 2025 research preview featured Maya and Miles, which Ars Technica called “eerily realistic.” Investor Sequoia reported that more than one million people used the voices during the first few weeks.
- Early 2025: Sesame released the Maya and Miles research preview.
- Closed beta: The company began testing a native iOS app.
- May 2026: The iOS preview opened in 39 countries.
- Later rollout: Sesame released an Android preview.
- General availability: Both apps launched with four agents and a new generation of underlying models.
The agents gain tools and schedules
Sesame says each agent now has its own “computer,” an agent-operated workspace that can browse the web and complete tasks through connected services. The release supports several practical capabilities:
- Web research: Investigate trips, products, and open-ended questions.
- Google access: Connect to Google integrations, including Gmail, Calendar, and Drive. OAuth grants scoped access without requiring users to share their passwords.
- MCP tools: Connect to third-party Model Context Protocol servers that authenticate through OAuth. MCP is a standard for exposing external tools and data sources to AI systems.
- Session continuity: Move an ongoing conversation between a spoken call and text.
- New outputs: Produce responses in multiple formats, including recorded voice notes.
Users can also create “skills” by describing a workflow in ordinary language, selecting connected apps, and specifying when it should run. A skill could summarize overnight email each morning or check the forecast before a recurring weekend hike.
Characters keep separate memories
Sesame treats its four agents as authored characters with individual voices, personalities, viewpoints, memories, and app permissions. The company says information shared with Maya remains outside Charlie’s memory, and each agent receives its own permission scope.
| Agent | Character profile |
|---|---|
| Maya | Warm and creative |
| Miles | Laid-back and sharp |
| Simone | Curious and intellectual |
| Charlie | Witty and warm |
Writers, producers, and actors work alongside Sesame’s researchers to shape the characters, giving personality design a formal role in model development.
Voice architecture cuts handoffs
Sesame’s Conversational Speech Model processes interleaved text and audio tokens as one sequence, reducing handoffs among separate transcription, language, and speech-synthesis components. The design preserves timing cues that help the agents respond to interruptions, use natural disfluencies, and maintain conversational pacing with less dead air.
The general-availability release uses a new generation of models, according to Sesame. The announcement does not provide model sizes, benchmark results, inference requirements, or details about which processing occurs on the device or in the cloud.
Security details remain sparse
Persistent memory and connections to email, calendars, and files create data-governance questions that the announcement does not resolve:
- How long conversation history and agent memory are retained
- How deletion works across backups and connected services
- Whether customer data may be used for model training
- Which actions require confirmation before an agent changes external data
- Whether users receive audit logs for scheduled jobs and connected tools
- Which MCP protocol versions, transports, and failure modes are supported
The glasses set the direction
Sesame plans to carry the same voice, memory, and connector stack into eyewear targeted for 2027. The company says the frames will be made in Japan and designed as everyday glasses with an assistant available through voice. It has not announced a price, final specifications, or an exact shipping date.
The mobile apps give Sesame time to refine its conversational models, tool integrations, permission system, and scheduled workflows before releasing hardware. They also test whether users will maintain ongoing relationships with named agents across calls, messages, and daily tasks.
Free rollout reaches 90 countries
| Platforms | iOS app and Android app |
|---|---|
| Geography | Roughly 90 countries |
| Price | Free during the rollout; later pricing has not been announced |
| Official language | English |
| Partial language support | French, Italian, German, Spanish, Chinese, Japanese, and Korean |
Competition turns on execution
OpenAI’s Advanced Voice Mode and Google’s Gemini Live already provide live spoken conversations, while wearable startups are developing assistants intended for use throughout the day. Sesame’s differentiation centers on low-latency turn-taking, authored characters, separated memory, and a direct path from mobile software to eyewear.
The mobile launch will test whether those qualities can support dependable research, connected-app actions, and scheduled work before Sesame’s hardware arrives.
Comments
