deepset is the company behind Haystack, an open-source framework for composing AI pipelines, and the Haystack Enterprise Platform, which adds build, test, deploy and govern capabilities. It targets agents and retrieval-augmented generation systems in which the team controls what the system knows, what it can do and how it reasons. The agent builder supports visual pipeline design and full-code development, models, tools and architectures can be mixed freely, and deployment options are cloud, self-hosted and VPC.
The pitch is sovereignty: most AI platforms decide infrastructure, models and data boundaries for you, and deepset positions Haystack as the opposite, with governance features including audit logs, role-based access control and data lineage, plus evaluation and testing. Named customers are the European Commission, Bosch and the German Federal Ministry for Research and Technology, which fits its focus on enterprises and governments in regulated sectors. It is headquartered in Berlin with a New York presence.
The open-source framework is free to use, while the Enterprise Platform has no published pricing and is reached through a demo request. Because Haystack is a toolkit rather than a finished assistant, expect engineering work: the value is in owning the pipeline, and the cost is in building and running it. Compare it with LangChain and LlamaIndex on evaluation and governance tooling rather than on connector counts.
