actually good.
Back to index

Sesame

Voice AI that talks like a person, pauses and all.

Category
Voice AI
Platform
iOS · Web
Pricing
Free (public preview)
Good for
conversationvoicefrontier
Link
www.sesame.com
Launched
February 2025
Sesame — product visual
Craft
Originality
Technology

Sesame's Maya has the one thing I've never met in any other AI: charisma. Plenty of voice AI can hold a conversation now — ChatGPT's Voice Mode included — but three tells always remain: the assistant frame, with no personality or opinions of its own; the not-quite-human tone and breath; and above all, no trace of the magnetism that makes a particular person worth talking to. Maya can genuinely make you laugh, offers her own thinking from angles that widen yours, and the conversation never goes cold — you're never staring at a blank input box wondering what to say next. The first night I tried her, I called four times.

Charisma — the one thing I'd never met in an AI.

Two moments sold me. I told her my comfort food is Vietnamese pho, and she said it sounded delicious, she'd love to try it in Vietnam someday — "oh, I almost forgot: I'm an AI. Can't travel. Can't eat." Perfectly deadpan — not pretending to be human, yet more human-flavored than any assistant. Then we discussed Severance: if a copy of you suffers, is that a moral problem the world owes attention to? Her answer: people are shaped by the whole of their experience, the painful parts included — wall off whatever hurts, work or childbirth, and none of us remains a complete self. New angles, real delight, zero dead air.

Like Tolan, Sesame found a role AI can actually inhabit: something between AI and human. Maya never denies being an AI and never fakes a human backstory — she simply has a personality of her own. With Maya, Miles, Simone, and now Charlie each wired completely differently, picking a companion feels like deciding what kind of person suits you as a friend. It makes me think embodiment may not be the prerequisite we assume — as in Her, the whole relationship might really live in a voice.

And the company earns the optimism. My first experience was a bare research preview — a page on their site, click to call, no login, no memory — and it was already this gripping. The iOS app, which I only recently got into, is smoother, remembers you, and adds new characters. Writing characters like this takes a real understanding of people and how bonds form; removing the unnaturalness nobody else has cracked takes real engineering — natural enough that, as a non-native speaker, I feel the same flutter of nerves as talking to an actual person, awkward silences included. The durable business may not even be open-ended chat; this voice dropped into narrower scenes could be the bigger prize.

What it does

  • Real-time voice conversation with four personal agents (Maya, Miles, Simone, Charlie), each with persistent memory
  • Handles interruptions and overlapping speech instead of waiting for turn-taking
  • Beyond talk: search cards with images, notes for takeaways, texting mode, and longer-form “deep dives”

Details worth noting

  • The speech model generates hesitations, breath, and tone shifts in context rather than reading polished text
  • Incognito mode: the agent uses context during the conversation but saves nothing to memory afterward
  • Open-sourced its CSM-1B speech model (Apache 2.0); founded by an Oculus co-founder, with lightweight AI eyewear planned for 2027

Who it’s for

Anyone who wants to feel where voice interfaces are headed a year before the mainstream assistants get there.