multi-strategy rag agent selection with automatic strategy routing
Implements three distinct RAG strategies (NaiveRAG, ChainOfRAG, DeepSearch) that can be selected via configuration or automatically routed based on query complexity. NaiveRAG performs single-pass retrieval-generation for simple queries; ChainOfRAG decomposes complex queries into sub-questions with iterative multi-hop reasoning and early stopping; DeepSearch executes parallel searches with LLM-based reranking and reflection loops for comprehensive research tasks. The agent selection is configuration-driven through the agent provider setting, enabling runtime strategy swapping without code changes.
Unique: Implements three distinct RAG agent classes (NaiveRAG, ChainOfRAG, DeepSearch) with pluggable selection via configuration, enabling strategy swapping without code changes. DeepSearch agent specifically combines parallel search with LLM-based reranking and reflection loops — a pattern optimized for reasoning models like DeepSeek-R1 and Grok-3.
vs alternatives: Offers more granular control over reasoning strategies than monolithic RAG systems; DeepSearch agent is specifically architected for reasoning models, whereas most RAG frameworks treat all LLMs equivalently
private data ingestion with multi-format file loading and web crawling
Provides pluggable file loader and web crawler implementations for ingesting diverse data sources into the vector database. Supports local file formats (PDF, text, markdown) and web content crawling through configurable loader and crawler provider classes. The offline_loading process orchestrates chunking, embedding generation via the configured embedding provider, and vector storage into Milvus or alternative vector databases. Data ingestion is decoupled from querying, enabling batch preprocessing of large document collections.
Unique: Implements pluggable loader and crawler provider classes that decouple data ingestion from querying, enabling batch preprocessing without blocking. The offline_loading orchestration layer handles chunking, embedding generation, and vector storage in a single pipeline, with provider selection managed through configuration.
vs alternatives: Separates ingestion from querying (unlike some monolithic RAG systems), enabling efficient batch processing; supports multiple file formats and crawlers through a unified provider interface without code changes
offline data loading pipeline with chunking and batch embedding generation
Implements the offline_loading process that orchestrates document ingestion, chunking, embedding generation, and vector storage. The pipeline loads documents using configured file loaders and web crawlers, chunks documents into fixed-size or semantic chunks, generates embeddings for each chunk using the configured embedding provider, and inserts embeddings into the vector database with metadata. This process is decoupled from query processing, enabling batch preprocessing of large document collections without blocking user queries. The pipeline is designed for one-time or periodic execution rather than real-time ingestion.
Unique: Implements a decoupled offline_loading pipeline that orchestrates document ingestion, chunking, embedding generation, and vector storage. The pipeline is designed for batch preprocessing, enabling efficient handling of large document collections without blocking query operations.
vs alternatives: Separation of offline loading from online querying enables better performance optimization; batch processing approach is more efficient than real-time ingestion for large collections
online query processing with context retrieval and llm-based answer generation
Implements the online_query process that retrieves relevant context from the vector database and generates answers using the configured LLM. The process encodes the user query as a vector embedding, searches the vector database for similar documents, constructs a prompt with retrieved context and the original query, and calls the LLM to generate an answer. The LLM has access to retrieved context, enabling it to provide grounded answers with citations. This process is optimized for low-latency query serving and can be executed repeatedly without modifying indexed data.
Unique: Implements online_query process that retrieves context from vector database and generates answers using the configured LLM. The process is optimized for low-latency serving and supports multiple RAG strategies (NaiveRAG, ChainOfRAG, DeepSearch) through pluggable agent selection.
vs alternatives: Unified query processing interface supports multiple RAG strategies without code changes; integration with vector database and LLM providers enables flexible technology stack selection
streaming response generation with token-by-token output
Implements streaming response generation that yields LLM output tokens one at a time rather than waiting for complete response generation. This capability is supported by LLM providers that implement streaming APIs (OpenAI, Anthropic, DeepSeek, etc.). Streaming enables real-time feedback to users, reduces perceived latency, and allows early termination if the user stops reading. The streaming interface is available through both the FastAPI web service (Server-Sent Events) and Python API (generator functions).
Unique: Implements streaming response generation through LLM provider streaming APIs, available via both Python API (generators) and FastAPI web service (Server-Sent Events). Enables real-time token-by-token output without waiting for complete generation.
vs alternatives: Streaming support reduces perceived latency compared to batch generation; available across multiple interfaces (Python API, web service) without code duplication
production deployment with docker containerization and kubernetes orchestration
Provides Docker containerization and Kubernetes deployment patterns for production deployment of DeepSearcher. The system can be containerized with all dependencies (Python, LLM clients, embedding libraries, vector database clients) and deployed as microservices. Kubernetes manifests enable horizontal scaling of query processing, load balancing across instances, and automatic failover. The FastAPI web service is designed for containerized deployment with health checks and graceful shutdown.
Unique: Provides Docker containerization and Kubernetes deployment patterns optimized for the FastAPI web service. Enables horizontal scaling of query processing and integration with managed vector database services (Zilliz Cloud).
vs alternatives: Kubernetes-native design enables horizontal scaling and high availability; integration with managed vector databases (Zilliz Cloud) simplifies infrastructure management
multi-provider llm abstraction with 17+ provider support
Provides a unified LLM provider interface that abstracts over 17+ language model providers including OpenAI, DeepSeek, Anthropic, Grok, Qwen, and local models. Each provider is implemented as a pluggable class (e.g., OpenAI, DeepSeek, AnthropicLLM, SiliconFlow, TogetherAI) with standardized method signatures for completion and streaming. Provider selection is configuration-driven via the llm_provider setting, enabling runtime swapping between cloud and local models without code changes. Supports both standard LLMs and specialized reasoning models (DeepSeek-R1, Grok-3).
Unique: Implements provider classes for 17+ LLM providers (OpenAI, DeepSeek, Anthropic, Grok, Qwen, SiliconFlow, TogetherAI, local models) with standardized method signatures, enabling configuration-driven provider swapping. Specialized support for reasoning models (DeepSeek-R1, Grok-3) that are optimized for multi-hop reasoning in RAG workflows.
vs alternatives: Broader provider coverage (17+) than most RAG frameworks; native support for reasoning models makes it better suited for deep research tasks than generic LLM abstraction layers
multi-provider embedding abstraction with 15+ embedding model support
Provides a unified embedding provider interface supporting 15+ embedding models from cloud providers (OpenAI, Cohere, Hugging Face) and local models (Sentence Transformers, Ollama). Each provider is implemented as a pluggable class with standardized embed() methods that return vector embeddings. Provider selection is configuration-driven via the embedding_provider setting, enabling runtime swapping between cloud and local embeddings. Embeddings are generated during offline_loading and used for semantic search during query processing.
Unique: Implements provider classes for 15+ embedding models (OpenAI, Cohere, Hugging Face, Sentence Transformers, Ollama) with standardized embed() interfaces. Supports both cloud and local embeddings through the same configuration interface, enabling privacy-preserving deployments.
vs alternatives: Broader embedding provider coverage than most RAG frameworks; unified interface for cloud and local embeddings makes it easier to migrate between privacy models without code changes
+6 more capabilities