Haystack
Open-source framework from Berlin for building RAG search, agents and chatbots in Python from combinable pipeline components.
What it does
Haystack is a Python framework under the Apache 2.0 licence for assembling AI applications from individual components. Pipelines connect data ingestion, retrieval, prompts and language models and can branch or loop. Typical uses are RAG search over your own documents, agents that call tools, and chatbots. Integrations include OpenAI, Anthropic, Mistral, Hugging Face, Elasticsearch, Pinecone and Weaviate. The companion project Hayhooks serves pipelines as HTTP endpoints or MCP servers. The Haystack Enterprise Platform adds a visual editor, testing, monitoring and access control.
Who it suits
Development teams who want to build search or assistants on company data themselves and control every step, rather than buying off-the-shelf software.
Data protection for businesses
The framework runs on your own infrastructure; where data goes depends on the models and databases you connect. Anonymous usage statistics, which exclude queries and document contents, are switched off with the environment variable HAYSTACK_TELEMETRY_ENABLED=False. The Enterprise Platform is available in the cloud or self-hosted, and since October 2026 also on STACKIT with data centres in Germany and Austria. deepset cites SOC 2 Type II and ISO 27001. According to its documentation, deepset does not use customer data to train or improve AI models. The data processing agreement is available via the Trust Center or your deepset account team.
Limits
You won’t get far without Python skills. The free Studio plan is limited to one user, 100 pipeline hours, 50 files and two development pipelines without production use. Enterprise pricing is only available on request.