Complex Adaptive Systems of AI

A neurosymbolic framework orchestrating LLMs and rule-based agents through formalized graph-based cognitive workflows.

Get Started View on GitHub

A New Approach to Cognitive Workflows

Imagine a cognitive platform that orchestrates various agents (rules, LLMs, etc.) based on workflows formalized as graphs for each decision process. This platform revolutionizes how we think about interactive systems.

Structure and Orchestration

analytics Logical Structure

  • Definition of graphs representing critical steps
  • Nodes embodying logic and processing
  • Automatic generation from structured knowledge
  • Support for both narrative and technical nodes
  • Integration with existing documentation (Obsidian)

settings Agent Orchestration

  • Contextual invocation of appropriate agents
  • Shared state for decision coherence
  • Blend of symbolic and neural approaches
  • Real-time agent coordination
  • Extensible agent architecture

Core Features

🔄

Graphical Workflows

Define or auto-generate decision graphs with nodes representing critical process steps and transition conditions. Each node can contain:

  • Business rules and conditions
  • LLM interaction points
  • User input handlers
  • State transitions
🤖

Agent System

A sophisticated multi-agent system featuring:

  • StoryGraph: Workflow and state transition management
  • RulesAgent: Game rules analysis and interpretation
  • DecisionAgent: Decision logic management
  • NarratorAgent: Content presentation and formatting
  • TraceAgent: Action history and character stats
💾

State Management

Immutable state system with:

  • Pydantic v2 validation
  • Complete action traceability
  • Decision tracking
  • Asynchronous event handling
  • In-memory caching

Multiple Applications

auto_stories

Interactive Storytelling

Create immersive narratives where:

  • Nodes describe key moments
  • LLM narrator writes immersive descriptions
  • Decision agent guides progression
  • Player choices influence the story
build

Industrial Diagnostics

Model verification protocols with:

  • Rule-based verification nodes
  • LLM interpretation of complex logs
  • Automated decision routing
  • Compliance tracking
account_tree

Branching Processes

Handle any branching workflow:

  • Training programs
  • Security audits
  • Intelligent checklists
  • Process automation

Technical Architecture

🔧 Core Technologies

  • FastAPI: REST API and WebSocket support
  • Pydantic v2: Data validation and serialization
  • LangGraph: Agent orchestration
  • OpenAI API (gpt-4o-mini): Language model
  • AsyncIO: Asynchronous processing

📦 Configuration Hierarchy

  • GameConfig: Main game configuration
  • AgentConfigs: Agent-specific settings
  • ManagerConfigs: Manager configurations
  • StorageConfig: Storage settings

🎮 Manager System

  • AgentManager: High-level coordinator
  • StateManager: Game state management
  • CacheManager: Memory cache handling
  • RulesManager: Game rules management
  • CharacterManager: Character handling

psychology Generative & Accessibility Features

Beyond manual workflow execution, Complex Adaptive Systems enables automatic graph generation from your existing documents

memory

Neurosymbolic assistant (LLM + rules)

history

Intuitive conversational interface

replay

Import from Obsidian notes

auto_awesome

Automatic workflow generation

visibility

Visual graph representation

Memory, History & Traceability

history

Comprehensive Logging

Every agent action is recorded with:

  • Request details
  • Response content
  • Decision justification
  • State changes
replay

Replay & Rollback

Advanced history features:

  • Decision history review
  • Workflow path replay
  • State rollback capability
  • Timeline visualization
description

Audit & Compliance

Complete decision journey tracking:

  • Detailed audit trails
  • Compliance documentation
  • Decision justifications
  • Process verification

Cognitive Workflow Architecture

A comprehensive framework combining:

graph

Graph Engine

Sophisticated graph processing with node management and transition logic

group

Agent System

Coordinated multi-agent system with specialized cognitive capabilities

memory

State Management

Robust state handling with immutability and validation

auto_awesome

Auto Generation

Intelligent workflow creation from structured documentation

accessibility

Accessibility

No-code interface for non-developers

extension

Extensibility

Flexible architecture for custom agents and workflows

Ready to Get Started?

Build your first cognitive workflow in minutes

Quick Start Guide View on GitHub