Txtai
txtai is an open-source Python framework for building AI-powered search and language model workflows. Developed by NeuML, the project is available at github.com/neuml/txtai, has approximately 12,600 GitHub stars, is Apache 2.0 licensed, and was last updated in May 2026.
The core primitive is an embeddings database — a unified index that combines dense vector search, sparse keyword (BM25) search, a graph network, and a relational SQL layer in one structure. On top of this, txtai provides pipelines (LLM prompts, summarisation, transcription, translation, question-answering), workflows that chain pipelines together, and autonomous agents that connect embeddings, pipelines, and external tools. A built-in FastAPI REST server and Model Context Protocol (MCP) API make it deployable as a microservice. Client bindings exist for JavaScript, Java, Rust, and Go.
Getting started requires three lines:
import txtai
embeddings = txtai.Embeddings()
embeddings.index(["document one", "document two"])
embeddings.search("query", 1)Placed in Languages & Frameworks / Assess / inner.
txtai addresses the assembly cost of building an AI search or RAG pipeline from individual components. A typical stack requires a separate vector store, a keyword search layer, an LLM client, a pipeline orchestrator, and an API layer. txtai replaces all of these with a single library, reducing the integration surface and time-to-working-prototype significantly.
The framework is the integrated alternative to the composable-primitives approach represented by other blips on this radar (DuckDB Vector Search, Search-based RAG, llm CLI). The tradeoff is breadth vs. lock-in: txtai covers more ground out of the box but ties the project to its abstractions. For projects where assembly cost is a genuine constraint — a solo developer or small team wanting AI search without building the stack piece by piece — txtai is the faster path.
12,600 stars, Apache 2.0 licence, and active commercial backing from NeuML place it firmly in Assess. Inner position reflects pip install txtai as the only entry point and 70+ example notebooks covering the full capability range.
Trial gate: txtai powering the search and retrieval layer of a real project, with at least hybrid search (vector + keyword) and one LLM pipeline in active use.
