Dynamic Agent is a Rust-based framework for building Retrieval-Augmented Generation (RAG) AI agents with multi-LLM and multi-vector store support, featuring WebSocket communication, caching, and flexible configuration.

Dynamic Agent

Dynamic Agent is a flexible and configurable AI agent framework built in Rust. It provides a foundation for creating Retrieval-Augmented Generation (RAG) agents that can interact with users over WebSockets, leveraging multiple LLM providers and vector stores.

Live Demo: Experience Dynamic Agent in action at thanon.dev/chat (powered by the Leptos Portfolio Admin frontend).

Key Features

• Multi-LLM Support: Integrates with various Large Language Model providers for chat completion, text embedding, and query generation.
  • Supported: Ollama, OpenAI, Anthropic, DeepSeek, DeepSeek, XAI, Groq.
  • Easily configurable via environment variables or CLI arguments.
• Streaming and Thinking Process: Supports both streaming responses and exposing the LLM's reasoning process.
  • Stream responses token by token for a responsive user experience.
  • Capture and stream the model's thinking process separately from the final response.
  • Control thinking display and duration based on client capabilities.
• GitHub Flavored Markdown Support: LLM responses can be formatted using GitHub Flavored Markdown, enabling rich text rendering on compatible frontends (e.g., code blocks, lists, bold/italics, tables).
• Multi-Vector Store Support: Leverages the vector-nexus crate to connect to different vector databases for RAG.
  • Supported: Redis, Qdrant, Chroma, Milvus, SurrealDB, Pinecone.
• Configurable RAG Pipeline:
  • Define agent behavior, intents, and prompt templates using local JSON files or Firebase Remote Config.
  • vector-nexus automatically detects your vector data structure, eliminating manual index_schema.json creation.
  • Supports LLM-driven intent classification and dynamic RAG query generation.
• **Dynamic P