My Agent Node

Advanced AI assistant with knowledge management, MCP tool integration, and multi-provider support.

Features

🚀 Streaming Architecture

• Real-time updates during execution
• Progressive output streaming
• Live progress tracking

🧠 Knowledge Management

• Short-term Memory: Stores recent interactions (last 20 entries)
• Long-term Memory: Stores important facts and knowledge (last 100 entries)
• Automatic context injection into system prompts
• Manual long-term memory saving

🔧 MCP Protocol Integration

• Tool calling via Model Context Protocol
• Support for Pieces and Flows as tools
• Built-in tools (mark as complete)

🤖 Multi-Provider Support

• OpenAI: GPT-3.5, GPT-4, and more
• Anthropic: Claude models
• Google: Gemini models
• Hugging Face: Popular open-source models (requires Hugging Face token)
  • Uses @huggingface/inference library
  • Supports streaming chat completion
  • Models: Llama 2, DialoGPT, Flan-T5, GPT-2, and more

🔄 Multi-Step Iteration

• Configurable max steps (default: 20)
• Automatic step counting
• Stop condition handling

Usage

Basic Usage

1. Select AI Model: Choose from available providers
2. Enter Prompt: Describe your task
3. Configure Tools: Add MCP tools if needed
4. Set Max Steps: Limit iterations (default: 20)
5. Enable Memory: Toggle short-term/long-term memory

Memory Management

Short-term Memory

• Automatically stores recent user prompts and assistant responses
• Used for conversation context
• Limited to last 20 interactions

Long-term Memory

• Manually save important information
• Persists across conversations
• Limited to last 100 entries
• Use "Save to Long-term Memory" checkbox and provide content

Example Workflow

1. User: "What is the capital of France?"
   → Saved to short-term memory

2. Assistant: "The capital of France is Paris."
   → Saved to short-term memory

3. User: "Remember that I prefer Python over JavaScript"
   → Save to long-term memory (manual)

4. User: "What programming language should I use?"
   → Agent uses long-term memory context
   → Responds: "Based on your preferences, you should use Python."

Configuration

Props

• Prompt: Task description
• AI Model: Provider and model selection
• MCP Tools: Array of available tools
• Structured Output: Define output schema
• Max Steps: Maximum iterations (default: 20)
• Use Memory: Enable/disable memory (default: true)
• Save to Long-term Memory: Save important info
• Long-term Memory Content: Content to save
• Hugging Face Token: Required for HF models

Memory Storage

Memory is stored in the project scope using the built-in store:

• Short-term: my_agent_short_term_memory
• Long-term: my_agent_long_term_memory

Limitations

• Short-term memory: 20 entries max
• Long-term memory: 100 entries max
• Hugging Face models require Hugging Face API token (get from HF Settings)
• Memory is project-scoped (shared across flows in same project)
• Hugging Face models may have different capabilities than OpenAI/Anthropic/Google models

Advanced Features

Custom System Prompts

The system prompt automatically includes:

• Current date/time
• Long-term knowledge base
• Recent conversation context (short-term)
• Core directives for tool usage

Tool Integration

• Pieces: Use any piece action as a tool
• Flows: Call other flows as tools
• Built-in: Mark as complete tool

Notes

• Memory persists across flow runs within the same project
• Memory is automatically trimmed to size limits
• Hugging Face support is experimental (may require custom implementation)
• Streaming updates provide real-time feedback