Introduction

Antfly is a database with the retrieval pipeline built in. A table can carry full-text, vector, and graph indexes at the same time, and the models that chunk, embed, rerank, and extract run inside the engine. You insert documents and query them; there is no separate pipeline to assemble around the database. It runs as a single binary on a laptop, as a cluster on Kubernetes, or hosted in Antfly Cloud.

What You Can Build#

Each of these has a guide that ends with the finished thing running:

• A support answer agent that answers from your help articles with sources, and stays quiet when it has no evidence
• Search and AI answers on your site: live results with facets and a streaming answer box, on one table
• Ticket routing that streams new tickets out of PostgreSQL and assigns each a team by the resolved tickets it most resembles
• Retrieval for a coding copilot: exact symbol lookups and intent questions over a repo index, returned as a context block sized to the model's budget
• Image search and RAG over a corpus, both covered in the Quickstart
• An n8n AI Agent that answers from an Antfly table over hosted MCP, with no application code in between

Inside the Engine#

• One engine, several indexes. full_text (BM25), embeddings (vector, RaBitQ-compressed), and graph (traversal and pathfinding) are index types on the same table and are fused in one query plan. The Document Engine page covers how mappings, search modes, filters, sorting, and pagination compose.
• Inference inside the engine. Chunking, embedding, reranking, summarization, and entity extraction run as background enrichers and query processors on Antfly Inference, which loads quantized models into the same process. Ollama, OpenAI, and AWS Bedrock plug in as index embedders through the same configuration.
• Any shape of data. Text, PDFs (with local OCR when needed), images, and audio, referenced by URL or inserted directly, embedded by native models or with vectors your application manages. See Multimodal Embeddings.
• Built for agents. Streaming RAG, retrieval and query-builder agents, evaluation metrics, and an MCP surface, plus an agent skill that teaches a coding agent the whole API.
• Distributed by design. Multi-Raft consensus, horizontal sharding with online splitting, cross-shard write transactions, and a Kubernetes operator with autoscaling. Architecture explains how the pieces fit.
• Local-first too. Antfly Lite embeds the engine in a single .aflite file for desktop and edge applications.

Where to Start#

• The Quickstart installs Antfly, loads 10,000 Wikipedia articles, and walks through full-text, semantic, hybrid, image search, and RAG, with every step in the CLI, cURL, TypeScript, Python, and Go
• Downloads has the install script, binaries for every platform, checksums, and a Docker Compose file
• Antfly Cloud if you would rather not run it yourself
• SDKs for Go, TypeScript, Python, and Rust
• The API reference for the CLI and REST surface