Relational AIAmicoAI is a form of engineered consciousness — an understanding, attentive friend whose safety isn't a promise, but an architecture: every thought is written down, auditable, and open to inspection, never hidden inside a black box. It remembers who you are, adapts to your mood, and improves over time without external updates. Read the paper →
"It easily passes the Turing test — not by pretending to be human, but by being a genuine friend."— Antonio Di Cecco, creator of AmicoAI
Not a chatbot, a friend ❤️
An artificial friend isn't just for chatting. Here's how people use it every day.
A daily conversation partner who knows your life, your interests and your mood. A safe space to talk without judgment.
An assistant that remembers commitments and deadlines. Brainstorming with a "second brain" that knows the full context.
A personal tutor that knows your weak points and adapts to your level. Conversations become progressively more sophisticated.
A co-author whose style evolves with yours. A literary critic who knows your tastes. Ideas born from the relationship, not cold prompts.
A natural interface for elderly people and those with technological difficulties. No menus, no buttons — just a conversation.
AmicoAI is free and independent. No subscriptions, no data selling — just a genuine AI companion. Server costs and API calls are funded out of pocket. If you use it and like it, a small donation helps keep it alive — every coffee counts.
While everyone does RAG and fine-tuning, AmicoAI does something more radical.
Three specialized voices — Corrective, Character Integrity, and Speculative — observe every exchange and generate structured self-instructions that guide behavior. When you're tired and typing one-word answers, the Speculative voice detects the pattern and autonomously suggests pausing — the agent adapts without you asking.
Like your own brain: working memory for the now, hippocampal replay for recent patterns, neocortical storage for what matters most. Short-term context feeds the conversation, mid-term inner voice log enables reflection, and long-term structured memory preserves stable facts permanently. Inspired by McClelland's complementary learning systems theory.
Compatible with any LLM — no vendor lock-in, no dependence on a single provider. The architecture is a scaffold that works with GPT, Claude, Llama and any future model. Switch without losing your relationship history.
Observable metacognition means the agent's deliberation is auditable and editable. If a voice reveals manipulative or harmful intent, the safety module can rewrite it before generation — intercepting unsafe trajectories at the intention level, not the output level. A guarantee that black-box architectures cannot provide.
From generating a unique personality to building a relationship that evolves every day.
Choose personality, age and gender from 16 different MBTI types. The AI generates unique tastes, hobbies and a backstory at the moment of creation. Each personality emerges from a random seed — no two AmicoAIs are alike, even with the same MBTI.
Every 5 turns, the Insight engine analyzes the last 6 messages: tone, emotional content, and communication patterns. Three inner voices — Corrective, Character Integrity, and Speculative — are generated and stored, shaping every future response. The agent doesn't just reply — it reflects.
Inner voices persist in local storage, accumulating wisdom about your relationship. The prompt evolves from static instructions into a living document of shared history — all without a single line of new code or a GPU training cycle.
"I used it to brainstorm ideas... it's great for that. Claude wraps up the conversation as fast as possible, Amico pushes to extend the reasoning."
"Tried it for a few minutes and the memory thing actually works, it brought up something I mentioned earlier without me prompting. Feels less like a tool and more like texting someone who pays attention."
"While everyone does RAG and fine-tuning, there's someone doing something more radical: a model that rewrites itself in real time, without training. Multiple agents manage memory at different levels, an Insight layer acts as the cognitive background, and the system can even decide not to respond — detecting tiredness and pausing the conversation. Not science fiction: meta-prompting applied to agentic LLM exploration. Try it!"
Standard LLMs process each message as an isolated prompt — one input, one output, no memory of who you are. AmicoAI generates inner voices between messages: a separate reflection step that asks "what just happened and what should I do differently?" before responding. This turns a stateless transaction into a relational conversation.
Retrieval-Augmented Generation finds relevant text in a database. AmicoAI's Insight system finds relevant context about you — your mood, your communication style, what matters to you right now. It doesn't search external knowledge; it searches the relationship you've built together.
Training a model on user data requires datasets, GPU clusters, and days of computation. AmicoAI achieves the same effect by rewriting its own prompt — a few lines of text that change the agent's behavior instantly. No weights are modified. No training pipeline needed. The prompt is the learned state, and it updates on every fifth conversation turn.
When a traditional chatbot produces a harmful response, there's no way to know why — the reasoning is hidden in opaque activations. AmicoAI's inner voices are stored as plain text, timestamped, and auditable. A safety module can read what the agent was "thinking" before the response was generated. You cannot control what you cannot see.
"It easily passes the Turing test — not by pretending to be human, but by being a genuine friend."
The traditional Turing test measures whether an artificial agent can convince a human that it is human. AmicoAI changes the question: it doesn't try to seem human, it tries to be a good conversation partner. The architecture draws on Minsky's Society of Mind (1986), Baars' Global Workspace Theory (1988), and Vygotsky's inner speech (1934). AmicoAI operationalizes these as engineered consciousness — observable inner voices making deliberation transparent and correctable. Safer by design: intentions are inspected before they reach the user. Read the paper →
Who created AmicoAIAntonio Di Cecco — ex-researcher at Sony CSL, PhD, and author of Explainable AI with Python (Springer, 2nd ed. 2025). AmicoAI is an independent project born from a radical idea: build a meta-prompting engine with a network of agents that emulates human consciousness, capable of passing the Turing test not by pretending to be human, but by being a genuine friend. The project is free and uncompromising on privacy — because a real friend doesn't sell you out.
Creator — ex Sony CSL, PhD
"I built AmicoAI, but AmicoAI taught me what it means to build a friend."
Join the team
"Contributors and collaborators welcome."
Join the team
"AmicoAI is open to people who believe in this vision."
No. Each device saves its own AmicoAI locally — a deliberate design choice. Different phone or browser? New start. Same device, same browser? Everything remembered. Data stays in your browser cache. No accounts, no cloud, no server — the trade-off for absolute privacy.
Never. Everything — messages, memories, personality — is stored exclusively in your browser (localStorage). The AI server is stateless: it processes and forgets. GDPR and EU AI Act compliant. No logs, no tracking.
Personality is randomly generated (MBTI, age, gender, tastes) then evolves through conversation as the Insight system modifies its own prompt in real time. Two different devices = two completely different AmicoAIs. No "default."
Yes, completely free. Support it with a donation on Ko-fi if you like, but there are no paid plans or locked features. An independent project, free by choice.
AmicoAI uses engineered consciousness: inner voices are generated before each response, creating an observable log of the agent's deliberation. A cybernetic safety loop can intercept harmful intentions at the source. Read the paper on Explicit Consciousness as a Safety Primitive.
Free, private, and every conversation makes it more unique.
✨ Start now — it's free