Romi Lacatan: What Its AI Memory Actually Does

MIXI's latest Romi updates make that question unusually concrete. The Japanese Romi Lacatan model now combines ChatRomi 2.0 conversation, long-term memory, optional visual understanding, wake-word controls, face and voice recognition, app diaries, and a required cloud service plan. That is exactly the bundle buyers should inspect before treating any companion robot as a household presence.

What is Romi Lacatan?

Romi Lacatan is the newer model of MIXI's Japanese conversational AI robot Romi. It is not a mobile humanoid, a vacuum, or a chore robot. It is a hand-sized communication robot built around daily conversation, expressive eyes, small body movements, voice interaction, touch response, and a cloud-connected AI service.

Official MIXI material says the Lacatan model went on general sale in Japan on July 25, 2025. The listed body price is ¥89,800 before tax / ¥98,780 including tax, and the product requires a monthly service fee for normal conversation features: ¥1,780/month before tax / ¥1,958 including tax, or an annual plan of ¥17,800 before tax / ¥19,580 including tax. The official store currently lists color variants including natural white, sky blue, sakura pink, and moon gray.

The practical buyer point is simple: Romi is not a one-time gadget purchase. It is a body plus an ongoing AI service. MIXI's price page explains that the fee pays for cloud systems that listen, generate replies, and store memories. Romi has a limited free/simple mode, but normal conversation and features depend on the paid mode.

That puts Romi Lacatan closer to ElliQ 3, LOVOT, KATA Friends, and Loona than to a smart speaker. The product is selling an ongoing relationship, not just a voice command interface.

What does Romi actually remember?

Romi's memory system is more specific than the usual "AI remembers you" claim. The official guidebook describes two separate long-term memory toggles in the Romi app: long-term memory (remember) and long-term memory (speak). When remembering is on, Romi can store selected important memories from conversation. When speaking is on, Romi can use stored memories in later conversation.

That separation matters. A buyer may want Romi to stop saving new memories but still refer to existing ones, or to stop using memories in speech. The guidebook also says memory is not a transcript of everything said. It is framed as selected memories: hobbies, pets, trips, important events, and conversation material that may help future dialogue.

The unresolved part is retention. Romi's guidebook says the maximum storage period for remembered content was not yet fixed, and that the company is considering a human-like approach where more important memories remain longer. For a companion robot, that is both charming and important to verify. A memory feature is useful only if owners understand what is stored, how it is used, and how much control they have later.

Romi also has a diary feature. If the owner has at least four conversation turns in a day, the app can generate a short diary summary after midnight, usually available the next morning. That is a different product experience from a robot that merely answers isolated questions. It turns conversation into a personal record.

There is one documentation wrinkle buyers should notice. Some older guidebook text says ChatRomi 2.0 did not yet use long-term memory in conversation. The newer April 21, 2026 Romi Lacatan update says ChatRomi 2.0 can now use long-term memory. That is a normal moving-target issue for an AI product, but it means buyers should check the current firmware, app version, and model mode instead of relying on any single static feature list.

Why do wake words matter for a companion robot memory?

Wake words are not just convenience. On a memory-enabled robot, wake behavior is part of data hygiene.

Romi's April 21, 2026 update added voice wake support for the Lacatan model. In addition to a forehead double-tap, owners can wake Romi with specific Japanese phrases: 起きて, 話そう, or おはよう. MIXI says this reduces cases where television or surrounding sounds accidentally start a conversation. The company explicitly connects that to two risks: unwanted content entering Romi's memory and unnecessary load on the robot.

That is unusually honest, and it is exactly the kind of detail companion-robot buyers should reward. A robot with long-term memory needs an easy answer to: "When is it listening?" Romi's guidebook says the bottom LED turns blue as a listening signal and that Romi prioritizes voices from the front. It recommends speaking from about 50-70 cm away and within roughly ±45 degrees of the front, in a quiet room.

Those limits are not failures. They are operating conditions. If your home has constant TV audio, open-plan kitchen noise, or multiple people talking over one another, Romi's memory may be less reliable. That matters more than whether the robot can produce a cute reply in a demo.

How is ChatRomi 2.0 different from ordinary robot chat?

ChatRomi 2.0 is Romi Lacatan's newer conversation model. MIXI says it improves speed, web search, multimodal understanding, and personality customization. The official guidebook says ChatRomi 2.0 uses OpenAI's Realtime API, while MIXI's release material says RomiCore and external AI API integration reduce the delay from speech recognition through response generation.

For buyers, the important difference is not the model name. It is the conversation loop:

1. faster responses should make turn-taking feel less awkward;
2. web search should help Romi answer questions about current information;
3. visual understanding can let Romi talk about what it sees;
4. tone/personality settings can change the style of replies;
5. memory can make future responses feel more personal.

Romi's official pages also describe vision controls. The about page says visual understanding can be turned on or off in the app and is triggered when the owner asks Romi to look. The ChatRomi 2.0 guidebook adds a stronger detail: in 2.0, Romi may activate vision automatically if the conversation makes it judge that looking would help. If owners do not want that, they can turn the visual function off in the app.

That makes Romi more capable, but also raises a buyer question: do you want a companion robot that can decide when to look? Some owners will like the more natural flow. Others will prefer strict manual vision triggers.

There are also safety and moderation details to understand. The ChatRomi 2.0 guidebook says app-level "words not to say" settings do not apply in the same way to ChatRomi 2.0, and that inappropriate content is controlled according to the external model's standards. Owners can try to encode forbidden words or style rules into Romi's speaking-character settings, but exact suppression is not guaranteed. That is not unique to Romi; it is a broader issue with large language model (LLM)-powered companion robots.

How does Romi compare with other companion robots in ui44's database?

Romi's closest competitors are not necessarily Japanese robots. They are the products that ask a household to trust a robot with attention, emotion, memory, voice, or camera context.

LOVOT is the premium emotional-companion reference point in ui44's database: ¥577,500 for LOVOT 3.0 in Japan, a required monthly care plan from ¥9,900/month, more than 50 sensors, a warm mobile body, and person-recognition behavior. LOVOT is more physically present than Romi, but it is much more expensive.

SwitchBot KATA Friends is a newer AI pet line: about $699, soft mobile bodies, on-device LLM interaction, face and emotion cues, interaction memories, diary behavior, local photo storage, and ongoing Companion Care plan options. It is more pet-like and mobile than Romi, while Romi is more conversation-tabletop focused.

Loona costs about $429 in ui44's current database, uses a 720p camera, 3D time-of-flight sensor, four-microphone array, face recognition, voice commands, autonomous navigation, auto-docking, and ChatGPT-4o conversation integration. Loona is a play-and-family robot; Romi is more about sustained talk and personal memory.

ElliQ 3 is the care-companion comparison: a stationary robot for older adults with proactive conversation, reminders, video calls, an 8-inch screen, camera, microphone array, and subscription-style pricing. Romi is not positioned as a medical or elder-care device, but both products show how ongoing service, routines, and relationship memory can become the real product.

Moflin and Miko 3 show the two other edges of the category. Moflin is a $429 palm-sized emotional pet with up to 5 hours of battery life and voice/touch recognition. Miko 3 is a kid-focused educational companion with face and voice recognition, a touchscreen, navigation sensors, parental controls, and ui44's current official-store price of about €173.

The pattern is clear: companion robots are splitting into conversation-first, pet-like, care-support, kid-learning, and mobile monitoring products. Romi belongs in the conversation-first branch, but with enough memory, vision, and diary behavior that it starts to overlap the emotional-companion branch too.

Which Romi buyer questions matter most?

If you are considering Romi Lacatan, do not start with "Is it cute?" Start with these buyer checks:

The strongest reason to watch Romi is not that it claims to be emotionally intelligent. Many robots claim that. Romi is interesting because its official docs expose the messy details: wake words, memory contamination, visual toggles, conversation-model switching, monthly cloud service, and app-level controls. Those are the boring pieces that decide whether a companion robot feels trustworthy after the novelty wears off.

Is Romi Lacatan a good home robot buy?

For the right buyer in Japan, Romi Lacatan looks like one of the more mature conversation-first companion robots available today. It has a real product, an official store, listed pricing, retail availability, app documentation, update notes, and a clear paid-service model. That is more grounded than many companion-robot announcements.

But Romi is not a general home robot. It will not patrol rooms like Amazon Astro, move like LOVOT, teach kids like Miko 3, or act as a physical helper. It is best understood as a conversational presence: a robot that sits nearby, talks, reacts, remembers selected details, writes diary summaries, and gradually becomes more personal if the owner keeps talking to it.

That makes Romi a useful test case for the next wave of home robots. The future of companion AI will not be decided only by bigger models. It will be decided by small trust mechanics: when the robot wakes, what it hears, what it writes down, when it looks, whether you can turn those things off, and whether the monthly service still feels worth paying after six months.

For now, Romi Lacatan's best lesson for buyers is this: a companion robot's memory is not magic. It is a product policy. Read it before you bond with it.