Skip to content

A stateful AI agent framework powered by the Cognitive Lattice to solve complex tasks with persistent memory and reliable tool orchestration.

License

Notifications You must be signed in to change notification settings

Sean-V-Dev/CognitiveLattice

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

34 Commits
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

CognitiveLattice: Intelligent AI Agent Framework

CognitiveLattice is a sophisticated AI agent framework that combines intelligent tool orchestration, persistent memory, and context-aware processing to create truly adaptive and capable AI assistants.

Rather than being just another LLM wrapper, CognitiveLattice implements a cognitive architecture that enables AI agents to:

  • Remember where they've been, what they're doing, and where they're going
  • Intelligently select and coordinate tools based on context
  • Execute autonomous web automation with intelligent planning
  • Maintain persistent session memory across interactions
  • Process documents with enhanced RAG (Retrieval-Augmented Generation)
  • Execute complex multi-step tasks with adaptive planning

🎬 Live Demo

Watch the system build a chipotle bowl with only a prompt and no hard coded pathing: AutonomousWebAgent4

(The gif is choppy to fit within the readme, to see the full playback at normal fidelity you can watch a mp4 at the link bellow)

AutonomousWebAgent.mp4

Watch the CognitiveLattice agent in action. This is not a scripted demo. It's a live demonstration of the Cognitive Lattice enabling a series of stateless API calls to be chained into a single, successful, multi-step task. The agent's ability to select the right tool and recall its own actions is entirely dynamic.

CognitiveLattice1


Key Features

Cognitive Lattice - Persistent Memory & Session Management

  • Hybrid State Management: Active task tracking + comprehensive event logging
  • Cross-Session Persistence: Session files can be loaded/resumed (user-selectable lattice loading coming soon)
  • Dynamic Context Extraction: Automatically builds relevant context from session history
  • Task Progress Tracking: Monitors multi-step task completion with step-by-step state
  • Model-Agnostic Memory: Lattice data works with any LLM - switch models without losing context

Intelligent Tool Management

  • LLM-Driven Tool Selection: Uses AI reasoning to choose appropriate tools
  • Generic Tool Architecture: Works with any tool, not hardcoded for specific domains
  • Contextual Parameter Extraction: Automatically extracts tool parameters from conversation
  • Tool Result Integration: Seamlessly integrates tool outputs into conversations

**Structured Task Execution **

  • Multi-Step Planning: Creates and executes complex task plans
  • Adaptive Step Management: Handles user input at any step, allows backtracking
  • Task Lock System: Maintains focus during active task execution
  • Progress Summarization: Provides comprehensive "what have we done so far" summaries

Autonomous Web Automation (v0.1)

** Comprehensive Test Suite Available**: See CognitiveLattice_Test_Suites_README.md for complete documentation of 100 test runs with full audit trails, performance metrics, and system validation.

Overview
CognitiveLattice's web automation system achieves 100% success rate across complex multi-step workflows using only natural language prompts—no hardcoded selectors or scripts. The system has been extensively tested with comprehensive documentation covering every decision, DOM interaction, and cognitive state transition.

Key Capabilities:

  • Intelligent Planning: Creates step-by-step plans for complex web tasks before execution
  • Cognitive Lattice Awareness: Avoids redundant steps by remembering previous actions
  • Smart Element Detection: Advanced DOM processing with context-aware element ranking
  • Real-time DOM Analysis: Adapts to dynamic content without hardcoded selectors
  • Progressive Candidate Disclosure: Provides top-10 most relevant selectors to AI for each step
  • Auto-Enter Functionality: Follows web standards (type in search fields, then press Enter)
  • Single-Step Execution: Precise step-by-step progression with full state tracking
  • Complete Audit Trails: Every prompt, response, DOM state, and decision is logged
  • Unified Architecture: Same cognitive lattice system as stepwise tasks

Validated Performance (100 Test Runs):

  • 100% Task Completion Rate across all complexity levels
  • 1,189 DOM Interactions executed successfully
  • 1,100+ Steps completed with zero failures
  • Average 3-5 minutes per complete order workflow
  • Cold Run Testing: Every test starts from scratch (no cached paths)

Test Coverage:

  • Simple orders (40 runs)
  • Complex customizations with multiple ingredients (60 runs)
  • Double protein configurations
  • Side items and drinks
  • Multi-item orders Test Suite Archive: CognitiveLattice_E2E_Acceptance_Suite_Tests.zip (65MB compressed, 730MB uncompressed)

Contains complete documentation for 100 test runs including cognitive lattice states, DOM debug files, AI decision logs, and audit trails.

For Full Documentation: See CognitiveLattice_Test_Suites_README.md for:

  • Detailed architecture explanation
  • File structure and navigation guide
  • Performance benchmarks and comparisons
  • Known limitations and scope
  • Instructions for reproducing tests

Advanced Document Processing (Architecture Complete - Reconnection Needed)

  • Enhanced RAG System: Sophisticated document analysis with external AI enhancement
  • Multi-Format Support: Handles various document types and structures
  • Semantic Search: Intelligent document querying with context awareness
  • Session-Based RAG Storage: Avoids JSON serialization issues with in-memory management

External API Integration

  • OpenAI Integration: Leverages GPT models for enhanced reasoning
  • Modular API Client: Easy to extend with other AI services
  • Error Handling & Fallbacks: Graceful degradation when external services unavailable
  • Token-Conscious Processing: Optimizes token usage while maintaining capability

Privacy & Security Architecture (Airgap Design)

  • Document Airgapping: Process documents locally without exposing content to external LLMs
  • Encryption-Ready: Built-in encoding/decoding system supports encrypted document transmission
  • Lattice Confidentiality: Session data can be encrypted before storage (implementation pending)
  • Model Independence: Switch between LLMs without exposing previous reasoning or context
  • Future-Proof Privacy: Maintains user confidentiality as AI models evolve