What it is
Haystack is an open-source Python framework for building LLM-powered applications: retrieval-augmented generation pipelines, semantic search, question answering and autonomous agents. Developers assemble pipelines from components (retrievers, generators, routers, tool-calling agents) that can be swapped without rewriting the surrounding code, and pipelines serialise to YAML for version control and deployment.
Interface and dependencies
Install with pip install haystack-ai. The project requires Python 3.10 or later (pyproject.toml). It is model- and vendor-agnostic: the Haystack homepage lists integrations with OpenAI, Anthropic, Mistral and Hugging Face, plus document stores/vector databases including Weaviate, Pinecone and Elasticsearch; the separate deepset.ai/integrations page adds Azure OpenAI and AWS Bedrock. A developer building a support-ticket RAG system, for example, would wire a document store, an embedding retriever and a generator component into a single pipeline object.
Where it fits and its limits
Haystack itself is the open-source orchestration layer; deepset also sells a separate commercial "Haystack Enterprise Platform" for managed/self-hosted deployment, logging and monitoring, so production observability and hosting are not included in the open-source package by default. Command-line and REST API deployment likewise aren't built into the core package; they're provided by the separate deepset-ai/hayhooks project. The docs introduction available at time of writing did not specify a full list of supported document stores in one place, so integration coverage should be checked per-connector before committing to a stack. The framework is Python-only; there is no first-party JavaScript/TypeScript SDK.
Licence
Released under the Apache License 2.0, per the LICENSE file in the deepset-ai/haystack GitHub repository.
