itch.io

Hi everyone!

I've been working on Adaptive Neural Companion, an experimental browser-based 3D AI Agent that combines AI conversation, persistent memory, personality, emotion, voice interaction and real-time physical motion inside a WebGL Cyber Room.

You can play the current Web Alpha here:

https://harry12345123.itch.io/adaptive-neural-companion-3d-ai-cyber-room

The idea behind the project is simple:

Instead of putting an AI inside a traditional chat box, what happens if the AI also has a real-time 3D body, persistent memory, emotional state and physical movement?

V3.4.5 — ADAPTIVE EMBEDDED CHAT

The latest v3.4.5 update focuses on making the AI chat and 3D companion work more naturally together inside the itch.io browser page.

The embedded layout automatically adapts when the game is running inside a smaller itch.io iframe.

The LIVE NEURAL LINK chat interface becomes more compact so it does not cover the 3D character.

When the game enters fullscreen mode, the complete chat interface and Cyber Room presentation are restored.

CURRENT FEATURES

• Browser-based WebGL2 3D rendering

• Animated GLB AI companion

• Adaptive embedded AI chat

• Streaming AI response architecture

• Fast Text-to-Motion

• Multi-action Motion Matrix sequences

• Persistent IndexedDB memory

• Personality persistence

• Continuous emotional state

• Voice input / Speech Recognition support

• Browser Text-to-Speech

• Emotion-driven 3D body language

• Procedural AI Cyber Room

• Emotion-reactive environment

• Holographic floor effects

• Runtime diagnostics

• Fullscreen adaptive layout

• Offline fallback mode

AI TEXT → MOTION → 3D ACTION

One of the main experiments in the project is converting natural-language instructions directly into physical 3D actions.

You can try commands such as:

"Turn left."

"Turn right."

"Walk forward."

"Walk forward two steps."

"Move closer to me."

"Look up."

"Look down."

"Give me a soft nod."

"Jump for joy."

"Spin around."

"Give me a gentle bow."

"Show curiosity."

"React with surprise."

You can also combine actions:

"Turn left, then walk forward."

"Turn right, walk forward two steps, then look up."

"Come closer to me, give me two nods, then listen quietly."

"Turn left, move forward, then jump for joy."

The current motion system supports ordered sequences of up to several physical actions.

FAST TEXT-TO-MOTION

Explicit motion commands can be detected before the complete AI response finishes generating.

The goal is to make the character feel more responsive:

USER MESSAGE

      ↓

MOTION INTENT

      ↓

MOTION MATRIX

      ↓

3D AVATAR ACTION

rather than:

USER MESSAGE

      ↓

WAIT FOR FULL AI RESPONSE

      ↓

ANIMATION

MEMORY

The browser version also includes persistent local memory using IndexedDB.

The companion can maintain information such as:

• Preferences

• Goals

• Important experiences

• Relationship context

• Recent conversation history

• Persistent personality state

The goal is to gradually move away from a completely stateless chatbot experience.

PERSONALITY + CONTINUOUS EMOTION

The companion also maintains a persistent personality and continuous emotional presentation.

Current state can include concepts such as:

• Warmth

• Energy

• Curiosity

• Confidence

• Calmness

• Engagement

• Familiarity

• Rapport

• Initiative

These values can influence how the companion talks and how expressive its physical behavior becomes.

VOICE

Supported browsers can also use voice interaction.

The current architecture includes:

• Microphone input

• Speech Recognition

• Live transcription

• Automatic voice sending

• Text-to-Speech

• Voice interruption / barge-in

Microphone access requires HTTPS and explicit user permission.

Browser support for Speech Recognition can vary, so desktop Chrome or Edge is currently recommended.

PUBLIC WEB BUILD

The public itch.io package intentionally does NOT contain a private AI API key.

The AI endpoint is left blank in the public package.

If no compatible external AI endpoint is configured, the project keeps an Offline Fallback so the Cyber Room, 3D avatar, Motion Matrix, memory and other local browser systems remain usable.

The goal is to avoid shipping private credentials inside public browser JavaScript.

THIS IS STILL AN EARLY VERSION

This Web Alpha is only the beginning of Adaptive Neural Companion.

I plan to continue developing the project with:

• More autonomous AI Agent behavior

• More 3D Motion Matrix actions

• Better AI-to-motion synchronization

• More emotional reactions

• Better long-term memory

• More personality evolution

• Improved voice interaction

• More Cyber Room interactions

• More 3D visual effects

• More avatar customization

• Better browser performance

• Deeper interaction between AI conversation and the 3D world

I'd really like future versions to be influenced by feedback from people actually using the project.

If you could add one new AI Agent feature, what would it be?

More autonomous movement?

Better memory?

More voice interaction?

More emotional behavior?

More 3D actions?

A more interactive Cyber Room?

Feel free to leave suggestions — I'm very open to ideas for future updates.

Thanks for checking out Adaptive Neural Companion!

Adaptive Neural Companion is now available on itch.io:

https://harry12345123.itch.io/a-native-windows-3d-ai-companion-with-memory-voice...https://www.youtube.com/watch?v=eGA02VmrHGY

Hi everyone!

I've been working on Adaptive Neural Companion, an experimental native Windows 3D AI Agent that combines AI conversation, long-term memory, personality, emotion, voice integration and real-time physical motion inside a 3D Cyber Room.

The idea is to explore something beyond a traditional chatbot:

What happens when an AI can not only talk to you, but also remember previous interactions and physically respond through a real-time 3D body?

CURRENT FEATURES

• Native Windows 3D runtime

• Animated 3D AI companion

• Fast Text-to-Motion

• Motion Matrix actions

• Multi-action movement sequences

• Long-term memory architecture

• Persistent personality

• Continuous emotional state

• Voice integration architecture

• Local AI support

• Automatic GPU routing

• Optional NVIDIA CUDA acceleration

• AMD / Intel / NVIDIA Universal fallback

AI → MOTION MATRIX → 3D ACTION

One of the main experiments in the project is converting natural-language instructions into physical 3D actions.

You can try prompts such as:

"Turn left."

"Walk forward two steps."

"Give me a soft nod."

"Jump for joy."

"Come closer to me."

"Turn right, walk forward two steps, then look up."

"Come closer to me, give me two nods, then listen quietly."

"Turn left, move forward, then jump for joy."

The system attempts to recognize the physical intent and route it into the existing Motion Matrix / MotionPlan system.

The goal is to reduce the feeling that you're simply talking to a static 3D model.

MEMORY + PERSONALITY + EMOTION

The project is also being designed around persistent interaction.

The companion can maintain systems for:

• Conversation context

• User preferences

• Goals

• Important experiences

• Relationship context

• Personality state

• Emotional presentation

These systems are intended to gradually influence how the companion communicates and physically behaves.

WINDOWS / GPU SUPPORT

The current release targets Windows 10 / Windows 11 x64.

NVIDIA GeForce / RTX GPUs are currently the primary optimization target and are recommended for the best experience.

The Universal edition also supports compatible AMD Radeon and Intel environments through the portable runtime path.

On supported NVIDIA systems, optional CUDA acceleration can be used for the Action Matrix backend.

FIRST-RUN NOTE

This release uses a Windows Self-Bootstrap player architecture.

Please extract the entire ZIP before launching:

AdaptiveNeuralCompanion.exe

On the first launch, the application may need to prepare the native Windows runtime. Internet access may be required if the required local toolchain is not already available.

If you launch the EXE and the game window does not appear immediately, wait for the first-run preparation to finish, close the application completely, and run it again.

If necessary, try launching it another 1–2 times.

Do not delete or rename the _anc_runtime folder.

THIS IS ONLY THE BEGINNING

This is currently an early version of the project, not the final version.

I plan to continue improving the Windows 3D AI Agent with:

• More autonomous AI behaviors

• More Motion Matrix actions

• More emotional reactions

• Better long-term memory

• Deeper personality evolution

• Better voice conversation

• More interactive Cyber Room systems

• More visual effects

• More avatar customization

• Better AI-to-motion synchronization

• Additional NVIDIA RTX / CUDA optimizations

I'd also really like the future direction to be influenced by people actually trying the project.

If there is an AI Agent feature you would like to see, please leave a comment.

What should the companion be able to do next?

More autonomous movement?

Better memory?

More emotions?

Voice conversation?

New 3D actions?

A more interactive Cyber Room?

I'm very open to suggestions.

Thanks for checking out Adaptive Neural Companion!

Thank you