@super_studio/ecforce-ai-agent-react

Provides a pre-built React component that renders a conversational interface for AI agent interactions, handling message rendering, user input capture, and real-time message streaming. The component integrates with the ecforce-ai-agent-server backend via HTTP/WebSocket protocols, managing UI state for chat history, loading states, and error boundaries without requiring custom chat UI implementation.

The ecforce-ai-agent-server component manages AI agent lifecycle, tool execution, and multi-turn conversation state on the backend. It handles agent initialization, function calling dispatch to external APIs, context management across conversation turns, and response streaming back to the React client via Server-Sent Events (SSE) or WebSocket, abstracting LLM provider complexity.

Implements Server-Sent Events (SSE) or WebSocket-based streaming to deliver AI agent responses incrementally to the React client, enabling real-time message rendering as tokens arrive rather than waiting for complete response buffering. The streaming layer handles connection lifecycle, error recovery, and message framing to ensure reliable delivery across network interruptions.

Enables AI agents to invoke external tools and APIs by parsing LLM function-calling outputs and dispatching them to registered tool handlers. The system validates tool schemas, manages tool execution context, and returns results back to the agent for continued reasoning, supporting both synchronous and asynchronous tool execution with error handling and timeout management.

Maintains conversation context across multiple agent-user exchanges, preserving message history, agent reasoning state, and tool execution results. The system manages context window optimization (summarization or truncation for long conversations), ensures consistent agent behavior across turns, and provides hooks for external persistence to databases or vector stores.

Abstracts underlying LLM providers (OpenAI, Anthropic, etc.) behind a unified interface, allowing agents to switch between models and providers without code changes. The system handles provider-specific API differences, token counting, and model-specific parameters (temperature, top_p, etc.), enabling flexible model selection at runtime or configuration time.

Implements error handling for agent execution failures including LLM API errors, tool execution failures, and network interruptions. The system provides retry logic with exponential backoff, error propagation to the client with user-friendly messages, and fallback mechanisms to gracefully degrade functionality when errors occur.

Provides a configuration system for defining agent behavior including system prompts, model selection, tool availability, temperature/sampling parameters, and execution constraints. Configuration can be defined at startup or dynamically at runtime, enabling different agent personalities and capabilities for different use cases without code changes.