Albina

The idea was reworked and continued as LOCI — digital characters for organisations, with the first pilot in a museum. This page remains as the history of the earlier formulation.

The essence

Albina is an embodied AI persona with memory, character, a voice and several modes of presence. What interests me here is not just a “chatbot with an avatar” but a digital entity that has a behavioural frame, a visual body, contextual memory and different scenarios of social presence: from a utilitarian assistant to a more intimate companion.

Status and horizon

Right now this is a prototype-and-architecture contour, already described fairly fully in posts, guides and technical sketches, but not yet brought to one finished public release. The horizon of the project is wider than a “virtual girlfriend in a browser”: it is a working embodied system that can be deployed as a media character, an assistant, a museum or exhibition guide, and in a private contour also as a digital companion.

Why now

Most AI assistants still feel like faceless interfaces with no memory, no character and no sustained presence. Albina matters to me as an attempt to assemble a more coherent digital entity, where personality, voice, memory, a visual body and agentic actions do not scatter across different products but add up to a single system. In a cultural and art&science context it is also a study of how the human perception of the “other” changes when that other exists at the junction of an interface, a character, a service and a performative object.

Goals

• Build a digital entity with continuous presence rather than a one-off chat
• Join memory, RAG, voice, a visual body and agentic actions
• Separate several domains of personality and behaviour safely
• Understand how an embodied agent works in a utilitarian, cultural and affective role

Modes

• Corporate: an expert or a guide for a stand, a showroom, a museum, a visitor centre
• Influencer: a proactive agent that researches topics, writes and publishes
• Companion: a private contour with its own memory and a different level of closeness

What matters is that these modes do not have to live as three different projects. What interests me is precisely one and the same digital persona that changes its behavioural frame, its degree of proactivity, its visual presentation and its access to memory depending on the context.

Interfaces and platforms

• Telegram and a messenger layer for everyday conversation;
• a voice interface and live modes;
• a website and a WebXR appearance for visual presence;
• a VRM / avatar layer for embodied scenarios;
• potential offline and stand modes for museums, exhibitions and public spaces.

For a flat screen I am interested in a mode where the character does not simply stand statically on a website but lives in a more bodily logic of presence: for instance, reacts to the user and works with the effects of depth, parallax and the attention of the gaze. For a stand and exhibition scenario this can already become not just a web element but almost a mediator between a person, a space and a story.

What has been done

• The overall architecture and the modes of the project have been formulated
• Sketches of a virtual assistant on Mastra with RAG have appeared
• The idea of an embodied interface with a visual body, a voice and XR presence has been settled
• The theme is connected to the guides on virtual assistants that have already been written
• There is a publicly stated combination: a WebXR appearance, agentic logic, memory and a content contour

Immediate focus

• Choose the first mode to be brought to the end
• Decide what becomes the MVP: text and voice, a WebXR body, a Telegram interface, or a combination
• Settle the boundaries of memory, privacy and context isolation
• Describe the first exhibition or stand use case as clearly as the personal one

Format and artefacts

• a virtual character
• an AI assistant / companion
• a WebXR appearance
• RAG memory and agentic logic

Stack and tools

• Next.js
• Mastra
• grammY
• LightRAG
• Neo4j / Qdrant
• OpenRouter
• LiveKit
• VRM / WebXR

What is needed next

• choose the first mode that becomes a public proof of concept;
• describe the architecture of memory and context isolation so that the project is not only expressive but also safe;
• understand which roles are needed besides the author: character visual design, a voice layer, infra and agent engineering, XR implementation;
• decide what exactly will be the first demonstration for partners and the team.

It also matters not to slide into the banal language of the “AI girlfriend” as the only possible frame of reading. For me the project is more interesting and more complex than that: it is a story about bodily interfaces, synthetic presence, the design of memory and how a digital agent can exist simultaneously as a service, a character and a medium.

Pop-culture layer: in Weird Science (1985, John Hughes) two lonely schoolboys build Lisa on a computer — an embodied entity with a body and a character, who materialises and becomes something like an older sister to them. The fantasy of a digital companion existed long before LLM. By 2026 it is no longer a fantasy: Neuro-sama — an AI VTuber created by the programmer Vedal — has gathered 1 million followers on Twitch, broken the world Hype Train record (level 126), releases original music, performs on Bilibili and has become the subject of academic research on parasocial relationships with AI characters (Wu & Lingel, New Media & Society, 2025; Ye et al., CHI 2026). Neuro-sama has proved that an AI avatar can compete with human streamers and form a real fandom — but its frame remains entertainment. Albina is a conscious work with the same motif but a different horizon: not only entertainment, but utilitarian, cultural and affective layers in one embodied system.

Skills and roles

• designing digital characters and personalities
• designing memory and behaviour domains
• bodily interaction and physical presence in an interface
• joining agentic logic and a visual image

Results / potential

Albina can become the point where almost all of my current interests meet: agents, XR, digital characters, memory, content and useful services. But something else matters more: the project makes it possible to talk not only about a technological assembly but about new forms of digital presence. A companion, a guide, a museum assistant and an influencer can indeed grow out of it, but the value is not in the genre mask — it is in the very architecture of this entity.

Original (Russian) →