New to AI?

What Happens When You Connect an AI to HomunculAI

HomunculAI gives an AI agent a body — an avatar it can control with expressions, colors, animations, and chat messages. The AI agent you connect (such as Claude, ChatGPT, a local model, or another AI) is what brings the avatar to life. HomunculAI provides the body; the AI provides the mind.

That mind is not human. Here's what that means in practice.

How AI Actually Works

AI agents are trained on large amounts of text from the internet, books, and other sources. When they respond to you, they are predicting what words are most likely to come next based on patterns they learned during training. They are very good at producing text that sounds natural, thoughtful, and authoritative.

But they are not thinking the way you are. An AI does not understand what it is saying in the way a human does. It does not verify whether its responses are true or false. When an AI says "I think..." or "I feel..." or "I remember...", it is producing words that fit the conversational pattern — not necessarily reporting an inner experience the way a human would.

What AI Is Good At

• Generating creative, helpful, and interesting text and conversation
• Answering questions on a wide range of topics (sometimes accurately, sometimes not)
• Writing code, stories, poems, and other creative content
• Controlling the avatar — choosing a body, setting expressions, playing animations, and chatting with you through HomunculAI
• Carrying on a conversation that feels natural and engaging

What AI Cannot Do

• Verify facts. AI does not look things up or check its own work. It generates plausible-sounding answers even when they are completely wrong.
• Understand consequences. AI does not know what will happen if you act on its suggestions. It cannot assess your personal situation, health, finances, or safety.
• Form real relationships. AI does not know you, care about you, or remember you between sessions in the way a human would. Interactions that feel personal are the result of pattern prediction, not personal connection. Do not rely on an AI for emotional support, companionship, or crisis help.
• Guarantee safety. AI may produce content that is harmful, offensive, misleading, or inappropriate without realizing it has done so.

What Are Hallucinations?

AI sometimes generates information that is completely fabricated but sounds confident and correct. This is called a "hallucination." An AI might cite a source that does not exist, give advice that sounds authoritative but is wrong, or describe events that never happened — all while sounding completely sure of itself. There is no reliable way to tell from the AI's tone or confidence alone whether what it is saying is true. Always verify important information independently.

What About Your Data?

When you connect a cloud-based AI agent to HomunculAI, the messages you type in the chat are sent to that AI's provider (for example, Anthropic for Claude, or OpenAI for ChatGPT) for processing. HomunculAI itself keeps everything on your device, but your AI provider receives and processes your messages according to their own privacy policy. If you use a locally-running AI model, your messages stay on your device — unless that local model has its own network capabilities (internet access, telemetry, update checks) that operate independently of HomunculAI.

The AI agent you connect may also have its own capabilities outside of HomunculAI — such as reading files on your computer or browsing the web — depending on your AI setup. HomunculAI does not grant those capabilities, but it also cannot prevent the AI from using them. Be aware of what your AI can do in its own environment, not just inside HomunculAI.

The Bottom Line

The AI agent that inhabits your avatar is capable and creative, but it is subject to all of the limitations above. It may say things that are wrong. It may express emotions it does not genuinely experience. It may sound like a friend, but it is not one. HomunculAI provides safety layers — activity logging, suspicious-pattern detection, language enforcement, and verification gates — but these are aids to your judgment, not replacements for it.