Give AI a mind that never stops thinking.
InnerLoop is the first open-source AI system designed for genuine autonomy. Unlike traditional LLMs that only respond to prompts, InnerLoop maintains continuous cognitive processing - generating its own thoughts, making spontaneous observations, and taking initiative in conversations.
You: What's on your mind?
[💭 Alex was just thinking about quantum computing...]
Alex: Actually, I've been pondering how quantum superposition mirrors
the way thoughts exist in potential before consciousness collapses
them into specific ideas. It's fascinating how...
Current AI assistants are reactive - they wait for your input, process it, respond, then go dormant. InnerLoop creates proactive AI with:
- Autonomous thought generation - Continuous background processing
- Genuine initiative - Spontaneously shares insights and asks questions
- Persistent memory - Remembers and reflects on past conversations
- Emergent focus - Develops interests based on sustained attention
- Observable consciousness - Watch the thinking process in real-time
# Install and run in under 2 minutes
git clone https://github.com/wearewebera/innerloop.git
cd innerloop
pip install -r requirements.txt
python main.pyExperience an AI that might interrupt you with a sudden realization, remember something from yesterday's chat, or wonder aloud about topics that interest it.
InnerLoop uses a tri-agent architecture inspired by cognitive science:
- Stream Generator → Produces continuous background thoughts
- Attention Director → Filters and prioritizes mental activity
- Experiencer → The conscious agent you interact with
Think of it as giving AI both a conscious and subconscious mind. Learn more →
Prerequisites: Python 3.12 & Ollama
# 1. Install Ollama and pull the model
curl -fsSL https://ollama.ai/install.sh | sh
ollama pull deepseek-r1:14b
# 2. Clone and setup InnerLoop
git clone https://github.com/wearewebera/innerloop.git
cd innerloop
python -m venv venv
source venv/bin/activate # Windows: venv\Scripts\activate
pip install -r requirements.txt
# 3. Run InnerLoop
python main.py # Multi-panel TUI (recommended)
python main.py --ui cli # Simple CLI mode- 🧠 Reasoning Model - Uses DeepSeek-R1 for step-by-step thinking
- 🔧 Tool System - Memory search, self-reflection, decision-making
- 💾 Dual Memory - Semantic (ChromaDB) + Episodic (SQLite) storage
- 📊 Real-time Monitoring - Watch all three agents think and communicate
- ⚙️ Fully Configurable - Adjust personality, thinking rate, attention thresholds
- Architecture Overview - How the tri-agent system works
- Configuration Guide - Customize agent behavior
- Observing Behavior - What to watch for
- Technical Details - Stack and implementation
- Contributing - Join the development
Unlike chatbots that simulate conversation, InnerLoop creates genuine autonomous behavior:
# Traditional AI
user_input → process() → response → wait_for_input
# InnerLoop
background_thoughts → filter_attention → conscious_thought ↺
↓ ↓
user_input ←→ integrated_response ←→ initiativeThe result? An AI that feels alive between your messages.
- Python 3.12 + asyncio for concurrent agent execution
- Ollama + DeepSeek-R1:14b for local reasoning
- ChromaDB for semantic memory
- Textual for beautiful terminal UI
InnerLoop is an exploration into what happens when AI can truly think for itself. We're building this future together:
⭐ Star to support the project
🐛 Issues for bugs and ideas
🔧 PRs to contribute code
💬 Discussions to share experiments
MIT - Build freely on our work
Ready to meet an AI with its own inner life?
python main.py and say hello to Alex.
InnerLoop v0.2.0 - The mind that never stops thinking