Infinity Overview

Infinity is an AI-native, open‑source database designed for LLM (Large Language Model) applications. It delivers ultra-fast hybrid search—combining dense/sparse vectors, tensor, and full-text—on structured and unstructured data. It supports RAG (Retrieval‑Augmented Generation) workflows while ensuring strong performance and scalability.

⚡️ Key Features

• Incredibly Fast
  ~0.1 ms query latency with 15 K+ QPS on million-row vector datasets.

• Hybrid Search (Vector + Full‑text + Tensor)
  Supports dense embedding, sparse embedding, full-text, tensor search, plus fused reranking (e.g., RRF, ColBERT).

• Rich Data Types & Indexes
  Handles strings, numerics, vectors (multi-vector too), tensors. Built‑in indexes: columnar, ANN (IVF/HNSW), inverted (BM25).

• AI‑Native Architecture
  Supports structured queries, vector retrieval, full-text search, and filtering in unified execution via a push‑based pipeline.

• Ease-of-Use
  Single-binary deployment, no external dependencies, and intuitive Python and HTTP APIs.

• ACID-Compliant Engine
  Columnar storage with full ACID support, zonemap/Bloomfilter, secondary indexes, background GC, WAL replay.

📌 Use Cases

• Building Retrieval-Augmented Generation (RAG) pipelines with deeply grounded knowledge.
• Powering semantic search across diverse data types (text, vectors, tensors).
• Driving LLM-based apps: question-answering, recommendation systems, conversational agents, and copilots.

📚 Learn More

• Infinity Homepage & Docs
• GitHub Repository – Apache‑2.0 | 3.9K ⭐
• Medium: AI‑native database intro
• PyPI: infinity‑sdk – for Python integration

🚀 Quickstart Snippet

from infinity import connect, NetworkAddress
inf = connect(NetworkAddress("127.0.0.1", 23817))
db = inf.get_database("default_db")
tbl = db.create_table("my_table", {
  "num": {"type": "integer"},
  "body": {"type": "varchar"},
  "vec": {"type": "vector,4,float"}
})
tbl.insert([{"num":1,"body":"hi","vec":[0.1,0.2,0.3,0.4]}])
res = tbl.output(["*"]).match_dense("vec",[0.1,0.2,0.3,0.4],"float","ip",5).to_pl()
print(res)