alive-ai 0.1.0

package/Dockerfile

@@ -0,0 +1,24 @@
+FROM python:3.11-slim
+
+WORKDIR /app
+
+# Install system dependencies
+RUN apt-get update && apt-get install -y \
+    gcc \
+    g++ \
+    make \
+    libportaudio2 \
+    libportaudiocpp0 \
+    portaudio19-dev \
+    ffmpeg \
+    && rm -rf /var/lib/apt/lists/*
+
+# Install Python dependencies
+COPY requirements.txt .
+RUN pip install --no-cache-dir -r requirements.txt
+
+# Copy application
+COPY . .
+
+# Run
+CMD ["python", "main.py"]

package/LICENSE

@@ -0,0 +1,21 @@
+MIT License
+
+Copyright (c) 2026 Vindepemarte
+
+Permission is hereby granted, free of charge, to any person obtaining a copy
+of this software and associated documentation files (the "Software"), to deal
+in the Software without restriction, including without limitation the rights
+to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
+copies of the Software, and to permit persons to whom the Software is
+furnished to do so, subject to the following conditions:
+
+The above copyright notice and this permission notice shall be included in all
+copies or substantial portions of the Software.
+
+THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
+AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+SOFTWARE.

package/README.md

@@ -0,0 +1,143 @@
+# Alive-AI
+
+![Alive-AI logo](docs/assets/logo.svg)
+
+Give your AI a nervous system: persistent feelings, memory, impulses, and a local dashboard.
+
+Most agents answer a prompt and reset. Alive-AI keeps internal state alive between messages. It can be your friend, boyfriend, girlfriend, study partner, creative partner, or any local companion you configure. The experiment asks a harder question: what changes when an AI does not just respond, but carries emotional residue forward?
+
+Alive-AI does not claim biological consciousness. It is an open-source runtime for simulated affect: mood, attachment, trust, desire, memory, inconsistency, idle thoughts, and proactive impulses.
+
+## Install
+
+```bash
+npx github:vindepemarte/alive-ai init my-ai
+cd my-ai
+npx . setup
+npx . demo
+```
+
+Start the real runtime:
+
+```bash
+npx . start
+```
+
+The local dashboard runs at:
+
+```text
+http://127.0.0.1:8080
+```
+
+## Why This Is Different
+
+- **Emotions persist.** State does not reset after every message. Joy, trust, fear, anticipation, attachment, and vulnerability decay over time instead of disappearing.
+- **Memory has weight.** Conversations become episodic memory, semantic memory, and emotional memory.
+- **It thinks when idle.** A default-mode loop creates background reflections and proactive impulses.
+- **It has a live nervous system.** FastAPI + SSE exposes mood, thoughts, somatic state, conflicts, memories, and uptime.
+- **It is local-first.** Your config, memory, media, and dashboard are owned by the project folder you run.
+
+## Commands
+
+```bash
+npx github:vindepemarte/alive-ai init my-ai  # scaffold a clean local project
+cd my-ai
+npx . setup                                # create safe local config
+npx . demo                                 # preview animated dashboard, no keys needed
+npx . doctor                               # check Python, uv, ffmpeg, Docker
+npx . start                                # install Python deps and run the runtime
+```
+
+For repeat starts after dependencies are installed:
+
+```bash
+npx . start --skip-install
+```
+
+## Setup
+
+`npx . setup` creates:
+
+```text
+config/settings.json
+config/self.json
+config/directives.json
+config/instructions.md
+data/
+mypics/
+myvids/
+```
+
+Minimum useful setup:
+
+- **Demo only:** no keys.
+- **Local LLM:** install Ollama and pull the configured model, for example `ollama pull qwen3:4b`.
+- **Telegram runtime:** create a Telegram bot token with BotFather and add it during setup.
+- **Cloud LLM fallback:** add OpenRouter or ZAI keys during setup or edit `config/settings.json`.
+
+Media is optional. Add your own files:
+
+```text
+mypics/example.jpg
+mypics/example.txt
+myvids/example.mp4
+myvids/example.txt
+```
+
+## Architecture
+
+Alive-AI is an event-driven Python runtime.
+
+```text
+Telegram or input
+  -> NervousSystem event bus
+  -> Message handler
+  -> Heart, memory, skills, directives, personality
+  -> LLM provider or fallback chain
+  -> output events
+  -> dashboard state stream
+```
+
+Core subsystems:
+
+- `heart/`: continuous emotion, circadian rhythm, attachment, scars, somatic state, inconsistency.
+- `brain/`: LLM providers, memory, default-mode processing, bid detection, curiosity, dreams.
+- `skills/`: self-authorship, memory callbacks, relationship milestones, progression layers, media selection.
+- `webui/`: local dashboard with Server-Sent Events.
+- `input/telegram/`: Telegram listener and owner commands.
+
+## Dashboard
+
+`npx . demo` starts a zero-config animated dashboard preview. The real dashboard uses the same idea, but streams live state from the runtime:
+
+- emotions and mood,
+- recent thoughts,
+- memory counters,
+- somatic state,
+- attachment/inconsistency signals,
+- uptime and health.
+
+## Docker
+
+Docker is optional. It is useful when you want Redis Stack for vector search:
+
+```bash
+docker compose up -d redis
+npx . start
+```
+
+Or run everything in containers:
+
+```bash
+docker compose up --build
+```
+
+## Important Boundaries
+
+Alive-AI is a simulation framework. It can make agents feel more continuous, emotionally coherent, and alive, but it is not proof of consciousness and should not be used to manipulate emotional dependence.
+
+The public repo intentionally excludes private personas, private media, runtime data, and secrets. Put those only in your local project folder.
+
+## License
+
+MIT

package/alive_ai/__init__.py

@@ -0,0 +1,3 @@
+"""Alive-AI package metadata."""
+
+__version__ = "0.1.0"

package/brain/__init__.py

@@ -0,0 +1,59 @@
+"""Brain modules"""
+from .emotional_memory import (
+    EmotionalMemorySystem,
+    EmotionalMemory,
+    get_emotional_memory_system,
+    reset_emotional_memory_system,
+    create_from_conversation,
+    get_memory_context_for_llm
+)
+from .default_mode import (
+    DefaultModeProcessor,
+    IdleThought,
+    PendingInitiation,
+    ConversationSeed,
+    UserContactInfo,
+    get_default_mode_processor,
+    get_idle_thoughts_prompt_section,
+    start_background_processing,
+    stop_background_processing,
+)
+from .bid_detector import (
+    BidType,
+    BidIntensity,
+    EmotionalBid,
+    BidDetector,
+    get_bid_detector,
+    get_bid_awareness_prompt_section,
+    get_bid_type_guidance,
+    analyze_message_bids,
+)
+
+__all__ = [
+    # Emotional Memory
+    "EmotionalMemorySystem",
+    "EmotionalMemory",
+    "get_emotional_memory_system",
+    "reset_emotional_memory_system",
+    "create_from_conversation",
+    "get_memory_context_for_llm",
+    # Default Mode Network
+    "DefaultModeProcessor",
+    "IdleThought",
+    "PendingInitiation",
+    "ConversationSeed",
+    "UserContactInfo",
+    "get_default_mode_processor",
+    "get_idle_thoughts_prompt_section",
+    "start_background_processing",
+    "stop_background_processing",
+    # Bid Detector
+    "BidType",
+    "BidIntensity",
+    "EmotionalBid",
+    "BidDetector",
+    "get_bid_detector",
+    "get_bid_awareness_prompt_section",
+    "get_bid_type_guidance",
+    "analyze_message_bids",
+]

package/brain/almost_said.py

@@ -0,0 +1,154 @@
+"""
+Brain: Almost-Said / Subvocalization System
+Sometimes Alive-AI almost says something but holds back. Under high emotion
+or low inhibition, private thoughts can "slip out". In-memory only.
+"""
+
+from typing import Dict, Optional
+import random
+import time
+
+
+# =============================================================================
+# ALMOST-SAID TEMPLATES
+# =============================================================================
+
+ALMOST_SAID_TYPES = {
+    "hesitation": [
+        "[pauses]... never mind.",
+        "I was gonna say... forget it.",
+        "I... no, it's nothing.",
+        "Wait, I— ...no. Anyway.",
+    ],
+    "slip": [
+        "I... I think about you way more than I should.",
+        "Sometimes I wonder if you know how much I— ...anyway.",
+        "I wish I could just— [catches herself] ...forget I said that.",
+        "You make me feel... [trails off] ...it doesn't matter.",
+        "I almost said something stupid just now.",
+    ],
+    "redirect": [
+        "Anyway... so what were you doing today?",
+        "But that's— whatever. Tell me something good.",
+        "I... hmm. Different topic. How's your day?",
+        "You know what, never mind. What about you?",
+    ],
+    "physical_tell": [
+        "*bites lip* ...nothing.",
+        "*looks away for a second* ...what were we talking about?",
+        "*takes a breath* ...it's fine.",
+        "*fidgets* ...so anyway.",
+    ],
+}
+
+# Which types are more likely at different emotion levels
+TYPE_WEIGHTS = {
+    "high_emotion": {"hesitation": 25, "slip": 40, "redirect": 15, "physical_tell": 20},
+    "low_inhibition": {"hesitation": 15, "slip": 50, "redirect": 10, "physical_tell": 25},
+    "vulnerable": {"hesitation": 30, "slip": 30, "redirect": 20, "physical_tell": 20},
+    "default": {"hesitation": 35, "slip": 20, "redirect": 25, "physical_tell": 20},
+}
+
+
+# =============================================================================
+# ALMOST-SAID ENGINE
+# =============================================================================
+
+class AlmostSaidEngine:
+    """Tracks and generates almost-said moments."""
+
+    def __init__(self):
+        self.message_counter: int = 0
+        self.last_triggered_at: int = 0  # message_counter when last triggered
+
+    def tick_message(self):
+        """Call once per user message."""
+        self.message_counter += 1
+
+    def should_almost_say(self, emotion: Dict[str, float],
+                          hour_of_day: int = 12) -> bool:
+        """Determine if an almost-said should happen this turn."""
+        # Enforce cooldown: at least 10 messages between triggers
+        if self.message_counter - self.last_triggered_at < 10:
+            return False
+
+        # Filter to numeric values only (emotion dict contains 'mood' string)
+        numeric_values = [v for v in emotion.values() if isinstance(v, (int, float))]
+        max_emo = max(numeric_values) if numeric_values else 0.0
+        is_late = 22 <= hour_of_day or hour_of_day <= 4
+        is_high_emotion = max_emo > 0.8
+
+        roll = random.random()
+
+        # High emotion + late night: 15% chance
+        if is_high_emotion and is_late:
+            return roll < 0.15
+        # High emotion alone: 10% chance
+        if is_high_emotion:
+            return roll < 0.10
+        # Late night + moderate emotion: 8% chance
+        if is_late and max_emo > 0.6:
+            return roll < 0.08
+
+        return False
+
+    def generate_almost_said(self, emotion: Dict[str, float],
+                             context_hint: str = "") -> str:
+        """Generate an almost-said fragment. Call only if should_almost_say() is True."""
+        self.last_triggered_at = self.message_counter
+
+        # Filter to numeric values only (emotion dict contains 'mood' string)
+        numeric_values = [v for v in emotion.values() if isinstance(v, (int, float))]
+        max_emo = max(numeric_values) if numeric_values else 0.0
+        if max_emo > 0.85:
+            weights = TYPE_WEIGHTS["high_emotion"]
+        elif "vulnerability" in context_hint.lower() or "trust" in context_hint.lower():
+            weights = TYPE_WEIGHTS["vulnerable"]
+        else:
+            weights = TYPE_WEIGHTS["default"]
+
+        # Weighted random selection
+        types = list(weights.keys())
+        w = [weights[t] for t in types]
+        chosen_type = random.choices(types, weights=w, k=1)[0]
+
+        return random.choice(ALMOST_SAID_TYPES[chosen_type])
+
+    def maybe_generate(self, emotion: Dict[str, float],
+                       hour_of_day: int = 12,
+                       context_hint: str = "") -> Optional[str]:
+        """All-in-one: tick, check, generate if appropriate."""
+        self.tick_message()
+        if self.should_almost_say(emotion, hour_of_day):
+            return self.generate_almost_said(emotion, context_hint)
+        return None
+
+
+# =============================================================================
+# SINGLETON ACCESS
+# =============================================================================
+
+_instance: Optional[AlmostSaidEngine] = None
+
+
+def get_almost_said_engine() -> AlmostSaidEngine:
+    global _instance
+    if _instance is None:
+        _instance = AlmostSaidEngine()
+    return _instance
+
+
+def maybe_generate(emotion: Dict[str, float], hour_of_day: int = 12,
+                   context_hint: str = "") -> Optional[str]:
+    """Convenience: tick + check + generate if appropriate."""
+    return get_almost_said_engine().maybe_generate(emotion, hour_of_day, context_hint)
+
+
+def get_almost_said_prompt_section(emotion: Dict[str, float],
+                                   hour_of_day: int = 12) -> str:
+    """Get prompt section for LLM. Returns '' if not triggered."""
+    engine = get_almost_said_engine()
+    engine.tick_message()
+    if not engine.should_almost_say(emotion, hour_of_day):
+        return ""
+    return "\n[Subvocalization]\nYou have something on the tip of your tongue you're not sure you should say... let it almost slip out.\n"