Antfly is a distributed hybrid search engine released by the Antfly DB team in March 2026. Just as GPT can understand any text, Antfly enables indexing and searching any type of data, including text, images, audio, and video, within a single engine. It is built on top of etcd's Raft consensus algorithm and CockroachDB's Pebble storage engine, and its core identity lies in hybrid search, which merges BM25 full-text search, dense vector similarity, and graph traversal into a single query. As suggested by its homepage slogan, "Self-learning Retrieval for Agents," it is designed to enable AI agents to find the precise needle in a vast haystack of data. Previously, implementing vector search required combining multiple systems, such as external APIs for embedding generation (e.g., OpenAI, Cohere), Elasticsearch for full-text search, and Neo4j for knowledge graphs, leading to multi-infrastructure complexity. Antfly integrates these three components into a single binary. Through its built-in inference engine, it can handle embedding generation, document chunking, and cross-encoder reranking locally, without requiring external API calls. Its multi-Raft architecture separates the metadata consensus group and the shard-specific storage consensus group, providing fault isolation, ensuring that a failure in a specific shard does not propagate to the entire cluster. It applies RaBitQ quantization to SPANN-based vector indexing and supports the HTTP/3 (QUIC) protocol, maintaining low latency even with large datasets. In life science research, Antfly is suitable for integrating heterogeneous data sources, such as papers, patents, clinical documents, and microscopic images, into a single search infrastructure. For example, PubMed paper text and pathology tissue images can be loaded into the same collection, and then text and images can be simultaneously hybrid searched using a natural language query such as "lung tissue patterns associated with EGFR mutations." By leveraging its built-in RAG agent functionality, search results can be automatically passed to an LLM for summarization, inference, and streaming of follow-up questions. Furthermore, a knowledge graph pipeline can be built to automatically extract and explore relationships between genes, diseases, and drugs through graph indexing, without requiring separate infrastructure. It seamlessly integrates with existing SQL workflows through a PostgreSQL extension (pgaf) and facilitates integration with the AI agent ecosystem through MCP server and A2A protocol support.