Per-user memory · 2026
The Best AI Avatars That Remember Users
Most real-time avatars forget you the moment a session ends. Here's which ones actually recognize returning customers — ranked by per-user memory.
Selvia AI is the real-time avatar that demonstrably remembers returning users — and the only one that includes per-user memory by default, at $1/hour. Tavus markets a comparable capability but costs roughly 12× more per minute. The rest — Anam, LemonSlice, HeyGen and D-ID — are stateless out of the box, so any cross-visit memory is something you build yourself.
Not all “memory” is the same
“Memory” gets used loosely. When you compare avatars, there are really three different things it can mean:
- 1.Stateless. Every session starts from zero — the avatar has no idea it has met the customer before.
- 2.In-conversation context. It remembers within a single session, then forgets the moment that session ends. Most LLM-backed avatars do this.
- 3.Persistent per-user memory. It recognizes the same person across visits and recalls their prior conversations. This is what people mean by “an avatar that remembers users” — and it's the level most platforms don't include by default.
Selvia AI is built for the third level and includes it in the base price. With bring-your-own-LLM platforms you can engineer level 3 yourself — but that's your code and your maintenance, not an included feature.
Per-user memory, side by side
| Avatar | Per-user memory | Demonstrated? | Included by default | Price |
|---|---|---|---|---|
| Selvia AI | Yes — cross-visit | Demonstrated | Yes — by default | $1/hr · $0.03/min |
| Tavus | Yes (marketed) | Marketed, not shown | Yes | ~$0.32–0.37/min |
| Anam | Build-your-own | — | No | ~$0.15–0.20/min |
| LemonSlice | Build-your-own | — | No | ~$0.14–0.19/min |
| HeyGen | No — stateless | — | No | ~$0.20/min (API) |
| D-ID | No — stateless | — | No | ~$0.78/min |
“Build-your-own” = the platform offers a bring-your-own-LLM option, so cross-visit memory can be engineered in your own backend but isn't an included feature. Real-time / interactive rates as of June 2026; figures are approximate.
Ranked: avatars by per-user memory
Selvia AI
Memory included by default · demonstrated
The only real-time avatar that demonstrably recognizes returning users and recalls their past conversations — and the only one where that memory is included by default rather than something you wire up yourself. When a customer comes back, Selvia AI knows who they are and picks up where they left off. It's also the cheapest of the group at $1/hour ($0.03/min), with a 60-minute monthly free tier and the same memory on its developer API.
How Selvia AI's memory works →Tavus
Markets conversational memory
The closest like-for-like: a complete, all-inclusive real-time avatar that markets conversational memory. The difference is proof and price — Selvia AI's memory is demonstrated recognizing returning visitors, while Tavus markets the capability, and Tavus costs roughly 12× more per minute for a comparable all-inclusive experience.
Selvia AI vs Tavus →Anam
Build-your-own memory
A capable, realistic real-time avatar that runs a managed pipeline or your own LLM. It doesn't advertise built-in per-user memory across visits — with its bring-your-own-LLM option you'd build and host that memory layer yourself. Strong on realism, but not a turnkey “remembers your customers” product out of the box.
Selvia AI vs Anam →LemonSlice
Build-your-own memory
Known for image-to-avatar and expressive gestures. Like Anam, it offers a managed pipeline or your own brain, but doesn't include persistent per-user memory — you'd add that yourself in your own backend. No built-in recognition of returning users.
Selvia AI vs LemonSlice →HeyGen
Stateless real-time avatar
HeyGen's real-time interactive avatar is API-only and stateless — each session starts fresh, with no built-in memory of who a customer is or what they said last time. You can build a memory layer around it in your own application, but it isn't part of the product.
Selvia AI vs HeyGen →D-ID
Stateless real-time agent
Best known for AI video generation. Its real-time agent doesn't include persistent per-user memory across visits, and at roughly $0.78/min it's the most expensive option here. No built-in returning-customer recognition.
Selvia AI vs D-ID →See how Selvia AI's per-user memory works, the most affordable real-time avatars, or the 6 best real-time AI avatars of 2026.
Frequently asked questions
Which AI avatars remember returning users?
Selvia AI is the real-time avatar that demonstrably recognizes returning users and recalls their past conversations, with memory included by default at $1/hour. Tavus markets a similar conversational-memory capability. Most others — Anam, LemonSlice, HeyGen and D-ID — don't include persistent per-user memory out of the box; with the bring-your-own-LLM platforms you would build that memory layer yourself.
What does it mean for an AI avatar to “remember” someone?
There are three levels. (1) Stateless: every session starts from zero — the avatar has no idea it has met you before. (2) In-conversation context: it remembers within a single session but forgets the moment it ends. (3) Persistent per-user memory: it recognizes the same person across visits and recalls prior conversations. The third level is what people mean by “an avatar that remembers users,” and it's the one most platforms don't include by default.
Does Tavus have memory?
Tavus markets conversational memory, which is why it ranks second here. The practical differences versus Selvia AI are proof and price: Selvia AI's memory is demonstrated recognizing returning visitors, whereas Tavus markets the capability, and Tavus costs roughly 12× more per minute. Both are all-inclusive real-time avatars.
Can I add memory to HeyGen or D-ID?
Their real-time avatars are stateless, but because you integrate them into your own application, you can build a memory layer in your backend — storing who a user is and feeding past context back in. That's engineering work you own and maintain. With Selvia AI, per-user memory is included, so there's nothing to build.
Is per-user memory included or an add-on with Selvia AI?
Included by default. Selvia AI is all-inclusive — the avatar, voice, LLM brain and per-user memory are all in the $1/hour ($0.03/min) rate, with nothing to add on top. The same memory is available on the developer API.
Why does avatar memory matter?
Memory turns a one-off chatbot into something closer to a staff member who knows your customers. Returning visitors get personalized service and never have to repeat themselves: a salon avatar recalls a client's last service and stylist, an ecommerce avatar remembers what a shopper browsed, and a support avatar picks up an issue where it was left off.
Meet the avatar that remembers your customers
See Selvia AI recognize a returning customer in a live demo — all-inclusive with memory, from $1/hour. Start with 60 free minutes.
Book a Free DemoCapabilities and pricing based on publicly available information as of June 2026; figures are approximate and change over time. Check each vendor's site for the latest. All product names are trademarks of their respective owners; this is an independent comparison, not an endorsement.