Haystack

Haystack is deepset's open-source Python framework for building RAG and LLM applications as composable, typed pipelines of components. The 2.x rewrite added a cleaner component model, cycles for agent loops, and broad integrations. The Apache-2.0 core is free to self-host, while deepset sells managed products (deepset Cloud/Studio) with hosting, a visual builder, and enterprise features on top.

Haystack lets you assemble AI applications from Components (single-purpose units like retrievers, rankers, and generators) wired together into Pipelines with typed, explicit connections. The 2.x line, a ground-up 2024 rewrite, added a cleaner component model, typed sockets, async-friendly execution, and support for cycles so pipelines can express agent loops as well as classic RAG flows. It integrates with many model providers and document/vector stores rather than locking you to one backend. The framework is Apache-2.0 licensed and free to self-host, which makes it attractive to teams that want production RAG without vendor lock-in. deepset, the company behind Haystack, monetizes through managed offerings, historically deepset Cloud and the deepset Studio/AI Platform, that add hosting, a visual builder, collaboration, and enterprise features on top of the open-source core. This gives a clear open-source-versus-managed split: build freely with Haystack, or pay deepset for a managed experience. Haystack is a mature, well-documented option that skews toward retrieval-heavy and search applications, reflecting deepset's NLP roots. It is more of an engineering framework than a turnkey app, so it rewards teams comfortable designing pipelines, and it continues to add agent capabilities (for example, state injection for tools in 2026 releases).