Capability · Framework — rag

txtai

txtai from NeuML is a compact embeddings-first framework. It bundles a local vector database (with optional backends like SQLite, DuckDB, Faiss, Postgres), full-text + graph + ANN hybrid search, RAG pipelines, workflows, agents, and a large catalogue of application recipes. It's a good choice when you want to run RAG and agentic search entirely inside your own process with minimal external infrastructure.

Framework facts

Category: rag
Language: Python
License: Apache 2.0
Repository: https://github.com/neuml/txtai

Install

pip install txtai

Quickstart

from txtai import Embeddings

embeddings = Embeddings(path='sentence-transformers/all-MiniLM-L6-v2')
embeddings.index([
    'US tops 5 million virus cases',
    'Canada\'s NHL returns to play',
    'The sky is blue today'
])
print(embeddings.search('sports', 2))

Alternatives

Chroma — pure embedded vector DB
Haystack — pipeline framework
LlamaIndex — RAG-focused framework
LangChain + FAISS — DIY alternative

Frequently asked questions

Is txtai a vector database or a framework?

Both. At its core is an embeddings database, but it wraps enough pipelines, workflows, and agent primitives that you can build complete RAG and agentic apps without pulling in another framework.

Does txtai need a GPU?

No. txtai runs CPU-only with reasonable defaults (MiniLM-style models). GPU accelerates embedding and inference if you have one, but all features work on a CPU-only laptop — useful for offline/edge deployments.

Sources

txtai — docs — accessed 2026-04-20
txtai — GitHub — accessed 2026-04-20