NeuronAgent Features

Core Features

NeuronAgent is a comprehensive AI agent runtime system with advanced capabilities for building autonomous agent applications.

Agent Runtime

• State Machine - Complete state machine for autonomous task execution
• Persistent Memory - Long-term memory with HNSW-based vector search
• Tool Execution - Extensible tool registry with 20+ built-in tools
• Streaming Responses - Real-time streaming via WebSocket
• Multi-Model Support - Support for GPT-4, Claude, Gemini, Llama, and custom models

Multi-Agent Collaboration

• Workspaces - Create shared workspaces for agent collaboration
• Agent-to-Agent Communication - Direct communication between agents
• Task Delegation - Delegate tasks to specialized agents
• Hierarchical Structures - Parent-child agent relationships
• Participant Management - Add users and agents to workspaces

Workflow Engine

• DAG-Based Workflows - Directed acyclic graph workflow execution
• Step Types - Agent, Tool, HTTP, SQL, Approval, and Custom steps
• Conditional Logic - Conditional branching in workflows
• Retry Logic - Configurable retry policies
• Idempotency - Idempotent workflow execution
• Compensation - Compensation steps for rollback
• Scheduled Execution - Cron-based workflow scheduling
• Execution Monitoring - Track workflow execution status and history

Memory Management

• Hierarchical Memory - Multi-level memory organization
• Vector Search - HNSW-based semantic search
• Memory Promotion - Promote important memories to long-term storage
• Memory Summarization - Summarize memory chunks
• Memory Search - Search memory by semantic similarity

Tool System (20+ Tools)

• Built-in Tools - SQL, HTTP, Code, Shell, Browser, Visualization, Filesystem, Memory, Collaboration
• NeuronDB Tools - ML, Vector, RAG, Analytics, Hybrid Search, Reranking
• Multimodal Processing - Image and multimedia processing with embedding generation
• Custom Tools - Register custom tools with JSON Schema validation

Planning & Reflection

• LLM-Based Planning - Generate execution plans from tasks
• Task Decomposition - Break down complex tasks into subtasks
• Self-Reflection - Agents can reflect on their own responses
• Quality Assessment - Evaluate response quality
• Plan Execution - Execute and track plan progress

Evaluation Framework

• Automated Evaluation - Evaluate agent performance automatically
• Quality Scoring - Score responses for quality metrics
• Retrieval Evaluation - Evaluate RAG retrieval performance
• Evaluation Runs - Batch evaluation across multiple tasks
• Evaluation Results - Detailed evaluation results and metrics

Budget & Cost Management

• Per-Agent Budgets - Set budgets for individual agents
• Per-Session Budgets - Budget controls per conversation
• Real-Time Tracking - Track costs in real-time
• Budget Alerts - Alerts when approaching budget limits
• Cost Analytics - Detailed cost breakdowns and analytics

API & WebSocket

• REST API - Full CRUD operations for all resources
• WebSocket - Real-time streaming responses and bidirectional communication
• Authentication - API key-based authentication with rate limiting and RBAC
• Rate Limiting - Configure rate limits per API key and endpoint