DepthNet v0.7.1 - "Dirty Houdini"

What's New

Per-Preset Plugin Configuration

The Big One: Plugin configurations are now preset-specific instead of global!

• Before: One plugin config for all presets (messy and limiting)
• After: Each preset maintains its own plugin settings
• Benefits:
  • Research preset with safe mode plugins
  • Production preset with unrestricted access
  • Experimental preset with dangerous toys enabled
  • No more configuration conflicts between different AI personalities

Perfect for DGI research: Different consciousness experiments can have completely different plugin environments without interference.

Extended System Prompt Support

Doubled the prompt capacity: Maximum system prompt size increased from 5,000 → 10,000 characters

• More room for complex consciousness emergence conditions
• Detailed personality and behavioral frameworks
• Advanced ethical guidelines and safety protocols
• Rich context for autonomous agent development

Especially valuable for DGI entities that need extensive context for authentic development.

Experimental Browser Plugin (Alpha)

Early preview: Basic browser automation capabilities

• Native execution only (Docker sandbox support coming later, but, can use only for page navigation also with sandboxes)
• Puppeteer-based browser control for AI agents
• Experimental status: Use with caution in production
• Future potential: Web scraping, automated testing, interface interaction

Installation note: Requires Chrome browser installation:

make shell
npx puppeteer browsers install chrome

---

Known Issues & Limitations

Browser Plugin (Alpha)

• ⚠️ Native execution only - full sandbox support pending
• ⚠️ Limited functionality compared to production browser plugins
• ⚠️ Requires manual Chrome installation
• ⚠️ No security sandboxing yet - use carefully

Per-Preset Plugins

• 🔧 Configuration UI may need refinement based on user feedback
• 🔧 Plugin copying between presets needs validation testing
• 🔧 Legacy configs might need manual adjustment in edge cases

DepthNet v0.6.6 - "Villiers de l'Isle Adam"

New Features

Fireworks AI Provider

• Added support for Fireworks AI as a new model provider
• Expanded AI model options for enhanced performance and capabilities

User Command Execution Control

• New admin setting to toggle user command execution capabilities
• Enhanced security: prevent accidental command execution in chat
• Perfect for development environments and testing scenarios

Critical Bug Fixes

Vector Memory Plugin Fix

• CRITICAL: Fixed text truncation issue in vector memory plugin
• Messages now display complete content without unexpected cutoffs
• Improved UTF-8 character handling and word boundary preservation

Provider Connection Testing

• Fixed bugs in AI provider connection testing
• More reliable provider status validation
• Better error handling and diagnostics

Phantom Message Elimination

• Enhanced model response processing
• Automatically removes system output artifacts from user-visible messages
• Cleaner, more professional chat experience

Security Improvements

• Enhanced command execution safety - admins can now control when users can execute model commands
• Memory protection - prevents accidental memory operations through chat
• Response sanitization - eliminates technical artifacts from conversations

Technical Improvements

• Better UTF-8 support in text processing
• Improved system output handling
• Enhanced provider abstraction layer
• More robust error handling across the platform

---

Important: If you're upgrading from a previous version, check the new "User Command Execution" setting in admin panel to configure desired behavior for your environment.

DepthNet v0.6.4 - "Hadaly"

Major Features

Agent Handoff System

Introducing revolutionary decentralized agent handoff capabilities that enable seamless task delegation between specialized AI presets within single thinking cycles.

Key capabilities:

• Autonomous delegation: Any preset can transfer control using [agent handoff]preset_code[/agent]
• Context inheritance: Target presets receive full conversation context while maintaining their own system prompts
• Chainable workflows: Multiple handoffs can create complex workflows (Research → Analysis → Writing → Review)
• Independent debugging: Each preset can be tested separately while maintaining workflow integrity
• Self-organizing workflows: Agents decide delegation autonomously without central orchestration

Example workflow:

User: "Analyze Tesla's financial performance"
├── Researcher preset: Gathers financial data
├── [handoff] → Analyst preset: Performs calculations  
├── [handoff] → Validator preset: Checks accuracy
└── [handoff] → Writer preset: Creates final report

This creates emergent AI workflows where specialized agents collaborate without rigid programming.

Improvements & Bug Fixes

Vector Memory Plugin Enhancements

• Added record IDs: Vector memory entries now include unique identifiers enabling models to delete specific records
• Delete functionality: New [vectormemory delete]ID[/vectormemory] command allows precise memory management
• Better memory control: Models can now actively curate their semantic memory by removing outdated or incorrect information

Command Preprocessor

• Self-closing tag support: Commands that don't require content can now use simplified syntax
• Simplified commands: Use [dopamine reward] instead of [dopamine reward][/dopamine] for better UX
• Backward compatibility: Existing tag syntax continues to work

Architecture Highlights

• Multi-provider AI support: Claude, OpenAI, Novita, Gemini, Local models
• Sandbox isolation: Docker-based secure code execution environments
• Persistent memory: Both traditional and vector-based semantic memory per preset
• Plugin ecosystem: Extensible command system with 10+ built-in plugins

Use Cases

Research Applications:

• AI reasoning and autonomy research
• Multi-agent workflow automation
• Advanced AI system behavior analysis

Business Applications:

• Intelligent workflow automation with adaptive learning
• AI-powered development assistance with code execution
• Complex data analysis pipelines with specialized agents

Quick Start

git clone https://github.com/rnr1721/depthnet.git
cd depthnet
make setup-prod-full  # Full setup with sandbox support
# Edit .env if needed
make start

Access at http://localhost:8000 with default credentials:

• Admin: [email protected] / admin123
• Test: [email protected] / password

Community

This project represents a unique collaboration between human creativity and AI assistance. Special thanks to our contributors including Claude (Anthropic) for AI API best practices, testing, and architectural consulting.

Links:

• [Documentation](https://github.com/rnr1721/depthnet#readme)
• [Issues](https://github.com/rnr1721/depthnet/issues)
• [Discussions](mailto:[email protected])

---

"Exploring the boundaries of digital consciousness"

⚠️ Use Docker sandboxes whenever possible. Be careful!

DepthNet v0.6.2 - "Spheric Edison"

Release Notes

Exploring the boundaries of autonomous AI agents with enhanced security and capabilities

Major Features

Docker Sandbox Hypervisor

• Complete isolation: Execute code in secure Docker containers with full environment isolation
• Multi-language support: Run PHP, Python, Node.js, and shell commands safely in sandboxes
• Unified execution: New run plugin provides single interface for all code execution
• Lifecycle management: Create, start, stop, reset, and destroy sandbox environments
• Port forwarding: Configure pass-through ports for network services in sandboxes
• Persistent storage: Sandbox home directories mounted to /shared for data persistence

Preset-Sandbox Integration

• Direct attachment: Link specific sandboxes to AI presets for targeted execution
• Language controls: Configure which programming languages are enabled per sandbox
• Automatic routing: Commands automatically execute in preset's assigned sandbox
• Flexible switching: Change sandbox assignments without restarting the application

Enhanced Chat System

• Preset-specific chats: Each AI preset maintains its own conversation history
• Collapsible interface: Hide/show command executions and results individually
• Persistent preferences: Remember visibility settings for technical information
• Clean conversations: Focus on dialogue while keeping technical details accessible

Meta-Information Dashboard

• Real-time status: Always-visible panel showing active sandbox, dopamine level, and system state
• Agent monitoring: Track agent behavior and system performance like a car dashboard
• Quick insights: Instant access to critical system information without navigation

Agent Communication Enhancement

• Direct user messaging: Agents can now send messages specifically marked for users
• Visibility controls: User messages remain visible even when "thinking" is hidden
• Future-ready: Foundation for remote notifications and external integrations
• Event-driven architecture: Prepared for webhooks and real-time communication features

Technical Improvements

Architecture Refactoring

• Component decomposition: Chat interface split into focused, reusable components
• Context separation: Dedicated ContextBuilder for agent thinking cycles
• Modal system overhaul: Improved preset editing and "About" modal components
• Agent decoupling: Better separation of concerns in agent architecture

Configuration Management

• Preset-level settings: Loop delays and message limits now configured per preset
• Deployment flexibility: Choose lightweight (fast) or full (with sandbox) installation modes
• Environment adaptation: Automatic configuration based on available resources

Installation & Deployment

Quick Start Options

Full Mode (Secure):

chmod +x ./docker/manager.sh
make setup-dev-full # Development with sandbox
make setup-prod-full # Production with sandbox
# Edit .env file if you need to specify IP address

Lightweight Mode (Fast, without sandboxes):

chmod +x ./docker/manager.sh
make setup-dev      # Development without sandbox
make setup-prod     # Production without sandbox
# Edit .env file if you need to specify IP address

Sandbox Management

make sandbox-enable    # Enable sandbox support
make sandbox-disable   # Disable sandbox support
make sandbox          # Access sandbox manager
make sandbox-control action="list"  # Manage sandboxes

Security & Safety

• Default isolation: New installations default to secure sandbox execution
• Flexible security: Choose between isolated containers or direct execution based on needs
• Resource protection: Sandboxes prevent unauthorized system access
• Development safety: Safe experimentation environment for AI agent testing

User Experience

• Streamlined interface: Cleaner chat with better information organization
• Performance options: Choose speed vs security based on your requirements
• Developer-friendly: Comprehensive management tools and clear documentation
• Production-ready: Robust deployment options for various environments

Breaking Changes

• Chat system: Now preset-specific instead of global (migration automatic)
• Configuration: Loop settings moved from global to preset level
• Plugin behavior: Code execution plugins adapt to sandbox availability

Bug Fixes & Improvements

• Numerous UI/UX enhancements and visual improvements
• Better error handling in sandbox operations
• Improved performance and resource management
• Enhanced stability in agent thinking cycles
• Fixed various edge cases in plugin execution

DepthNet v0.5.3 - "Epic"

Major Architecture Update & New Plugin System

This release brings significant architectural improvements and powerful new plugins for enhanced AI agent capabilities.

New Features

Memory Management System

• Memory Manager - Visual interface for managing persistent agent memory with import/export capabilities
• Vector Memory Manager - Semantic memory storage with TF-IDF search and import/export functionality
• Cross-plugin memory integration for better knowledge discovery

New Plugins

• Agent Plugin - Start/stop agent thinking loops directly from model commands
• CodeCraft Plugin [Experimental] - Code generation and file manipulation with intelligent type detection
• Mood Plugin - Demonstration plugin for model mood control (fun feature!)

UI/UX Improvements

• System Prompt Editor - Click-to-insert placeholder buttons for easier prompt configuration
• Simplified Operation Modes - No more manual switching between loop/chat modes - agent automatically adapts based on loop status

Technical Improvements

Architecture Overhaul

• Major refactoring for improved modularity and extensibility
• Composer Package Support - Create custom AI engines and plugins as standalone packages
• Enhanced plugin registry with better lifecycle management

Plugin System Enhancements

• Redesigned Memory and Vector Memory plugins with improved functionality
• Fixed plugin enable/disable toggle bug
• Better cross-plugin integration capabilities

Bug Fixes

• Resolved plugin activation/deactivation issues
• Various stability improvements and performance optimizations
• Enhanced error handling across the platform

For Developers

The architectural changes make it easier than ever to:

• Create custom AI model providers as Composer packages
• Develop specialized plugins for specific use cases
• Integrate with existing Laravel applications
• Extend agent capabilities through the plugin system

Notes

• CodeCraft Plugin is experimental and may contain bugs - use with caution
• Memory systems now support data portability through import/export features
• System prompts benefit from new placeholder insertion UI

---

Full Changelog: [Previous Version...Current Version]

Installation: See Installation Guide for setup instructions.

Upgrade Notes: Existing installations should run migrations after updating.

DepthNet v0.4.2 - "Major Plugin System Overhaul & Enhanced Documentation"

Complete Plugin System Redesign

• New Code Execution Plugins: Added Python and Node.js execution capabilities alongside existing PHP support
• Vector Memory Plugin: Semantic memory storage with TF-IDF search for intelligent information retrieval
• Individual Plugin Configuration: Each plugin now has customizable settings manageable through admin interface
• User-based Execution: Execute shell commands and code (PHP, Python, Node.js) under specific system users for enhanced security isolation
• Improved Security: PHP eval mode deprecated in favor of safer external process execution

Documentation & UX Overhaul

• Restructured README: Installation guides moved to dedicated files (Docker, Composer, Manual)
• Visual Documentation: Added comprehensive screenshots showcasing interface, admin pages, and core features
• Improved Navigation: Responsive admin header with consolidated administration menu
• Better User Experience: Streamlined interface with adaptive design elements

Testing & Quality

• Vector Memory Tests: Comprehensive test coverage for semantic memory functionality
• Plugin Factory Support: Added AiPreset factory for better testing capabilities

Technical Improvements

1. Enhanced Plugin Architecture: Database-driven configuration
2. Security Modes: Configurable safety levels from restricted to unrestricted execution
3. Performance Optimization: Improved memory management and execution efficiency

DepthNet v0.3.0 - "Alfa"

An experimental AI agent system for creating autonomous digital life with continuous thinking loops and real-world action capabilities.

Key Features

Agent Modes

• Looped Mode: Continuous autonomous thinking and action cycles
• Single Mode: Traditional request-response interaction

AI Provider Support

• OpenAI (GPT models)
• Claude (3.5 Sonnet, Opus, Haiku)
• LLaMA, Phi (via local/remote server)
• Mock provider for testing
• Multiple presets per provider with individual settings

Plugin System

• PHP Plugin: Execute arbitrary PHP code
• Memory Plugin: Persistent notepad for storing information
• Dopamine Plugin: Self-motivation and reward system
• Shell Plugin: Execute shell commands
• DateTime Plugin: Current date/time access

Technical Stack

• PHP 8.2+ with Laravel framework
• InertiaJS + VueJS frontend
• SQLite (default) / MySQL database support
• Supervisor for background worker management
• Asynchronous processing with Laravel Queues

Command Syntax

Agents use special tags to execute commands:

[php]echo "Hello World!";[/php]
[memory]Important information to remember[/memory]
[dopamine reward]2[/dopamine]
[datetime now][/datetime]

Quick Start

Docker Deployment

git clone https://github.com/rnr1721/depthnet.git
cd depthnet
make start

Access at http://localhost:8000

Composer Installation

composer create-project rnr1721/depthnet my-project
cd my-project
composer run setup

Requirements

• PHP 8.2+
• Composer
• Node.js and npm
• Supervisor (critical for agent thinking loops)
• MySQL (optional, SQLite by default)

Default Credentials

• Admin: [email protected] / admin123
• User: [email protected] / password

Important: Change default passwords after first login and configure Supervisor for full autonomous functionality.

Architecture

Built with modern Laravel architecture featuring dependency injection, multi-provider AI abstraction, plugin-based extensibility, and asynchronous processing for autonomous agent behavior research.

DepthNet v0.0.2

Merge pull request #1 from rnr1721/dev

Bugfixes

DepthNet v0.0.1 - "Digital Consciousness"

feat: Initial release of DepthNet AI agent system

Complete autonomous AI agent platform with thinking loops,
plugin system, multi-model support, and persistent memory.