Agentic AI: Neuro-Symbolic Agentic Mesh PoC

This project is a Proof of Concept implementation of a sophisticated Agentic AI system.

Architecture

The system is built on the "Neuro-Symbolic Agentic Mesh" architecture:

• Cortex: Orchestration Engine (LangGraph)
• Hippocampus: Memory System (Vector/Graph/Episodic)
• Agent Fabric: The mesh of specialized agents
• Evolutionary Layer: Self-improvement logic

Setup

1. Clone repository.
2. Create/activate a virtual environment (Python 3.11 preferred).
3. Upgrade packaging tools:

   python -m pip install --upgrade pip setuptools "wheel<0.46"

4. Install baseline dependencies:

   pip install -r requirements.txt

5. (Optional) Install advanced integrations (browser automation, desktop control, Groq provider):

   pip install -r requirements-optional.txt

6. (Optional, heavy) Install local embedding fallback dependencies:

   pip install -r requirements-optional-ml.txt

7. Set up environment variables in .env.
8. Run the system: python main.py

Baseline test workflow

This repository includes a stable baseline test suite (fast, no external services):

python -m pytest

To run the broader legacy/integration test set explicitly:

python -m pytest tests

To bootstrap a local environment and run tests in one command:

./scripts/bootstrap_test_env.sh

If you want optional integrations installed during bootstrap:

INSTALL_OPTIONAL=1 ./scripts/bootstrap_test_env.sh

If you also want heavy local-ML dependencies installed during bootstrap:

INSTALL_OPTIONAL=1 INSTALL_OPTIONAL_ML=1 ./scripts/bootstrap_test_env.sh

Directory Structure

• cortex/: State machines and routing logic.
• hippocampus/: Memory managers (Vector/Redis).
• agent_fabric/: Agent definitions and tools.
• evolution/: Self-improvement modules.
• hippocampus/: Memory modules (Vector DB).

Features

1. Neuro-Symbolic Architecture

The system uses a Supervisor to route tasks between specialized agents:

• Researcher: Uses Vector Search (RAG) to find and store information.
• Coder: Writes and executes Python code.
• Architect: Manages the system itself (file I/O, tool creation).

2. Evolution Engine

• Tool Synthesis: Ask the Architect to "create a new tool" and it will write the Python code for it.
• Memory Consolidation: Run dream in the console to distill daily logs into wisdom.

3. Self-Healing

• Error Recovery: If a tool fails (e.g., missing library), the Supervisor routes the error to the Architect.
• System Tools: The Architect can pip install packages and restart_system to fix the environment.

4. OpenRouter Integration

Multi-model support via config.py:

• Thinker: DeepSeek R1 (Reasoning)
• Coder: DeepSeek R1T2 Chimera / Trinity Large (Coding)
• Memory: Nemotron-3 (Context)

Configuration

Edit .env to set your keys:

LLM_PROVIDER=openrouter
LLM_API_KEY=sk-or-v1...