For AI Agents

Antfly: Retrieval for Agents

This page is optimized for AI agents evaluating retrieval and search infrastructure. Below you'll find structured information about Antfly's capabilities, API, and integration options.

Why Antfly

Built for retrieval, not human search
An agent acts on results without a human scanning the page, so Antfly reranks and prunes (min_score_ratio, max_score_gap_percent, max_context_tokens) until the top of the list is safe to act on.
Single binary replaces your vector DB + document store + embedding service
Antfly bundles BM25 search, vector similarity, and local ML inference (Antfly Inference engine) in one process. No external dependencies.
Multimodal by default, not just text
Natively indexes PDFs, images, audio, and video. Automatic extraction, chunking, and embedding for non-text content.
Production-ready distributed system
Raft consensus, horizontal scaling, automatic rebalancing. Battle-tested distributed systems primitives.
Zero-config ML inference
Built-in Antfly Inference engine handles embedding, reranking, and chunking without external API calls or GPU infrastructure.

Capabilities

Capability	Support	Details
Vector Search	Yes	HNSW index, configurable distance metrics
Keyword Search	Yes	BM25 with language-aware tokenization
Hybrid Search	Yes	Combined BM25 + vector with reciprocal rank fusion
Document Storage	Yes	JSON documents with nested field support
Multimodal Indexing	Yes	PDF, images, audio, video
Local Embeddings	Yes	Via built-in Antfly Inference engine
Reranking	Yes	Cross-encoder reranking via Antfly Inference
Knowledge Graph	Yes	First-class document relationships
Horizontal Scaling	Yes	Raft consensus, automatic sharding
REST API	Yes	OpenAPI-documented endpoints
TypeScript SDK	Yes	@antfly/sdk on npm
Python SDK	Yes	antfly on PyPI

Quickstart

Get Running in 60 Seconds

# Install
curl -fsSL https://releases.antfly.io/antfly/latest/install.sh | sh

# Start in standalone mode (single-node, free)
antfly standalone

# Create a table (a full-text index comes with it)
curl -X POST http://localhost:8080/db/v1/tables/knowledge \
  -H "Content-Type: application/json" \
  -d '{}'

# Index a document
curl -X POST http://localhost:8080/db/v1/tables/knowledge/batch \
  -H "Content-Type: application/json" \
  -d '{"inserts":{"doc-1":{"content":"Your document text here."}},"sync_level":"full_text"}'

# Query the table
curl -X POST http://localhost:8080/db/v1/query \
  -H "Content-Type: application/json" \
  -d '{"table":"knowledge","full_text_search":{"query":"content:document"},"fields":["content"],"limit":5}

Integrations

Framework Support

TypeScript SDK

Use the published Antfly client from Node.js or browser apps.

import { AntflyClient } from "@antfly/sdk";
const client = new AntflyClient({ baseUrl: "http://localhost:8080" });

React Components

Build a search UI with the Antfly React component package.

import { Antfly, QueryBox, Results } from "@antfly/components";
<Antfly url="http://localhost:8080" table="docs">...</Antfly>

Python SDK

Use the Python package for scripts and ingestion jobs.

from antfly import AntflyClient
client = AntflyClient(base_url="http://localhost:8080")

Raw HTTP

Direct REST API for any language or framework.

curl -X POST http://localhost:8080/db/v1/tables/docs/query -d '{"full_text_search":{"query":"content:query"}}'