What is this?
An open-source AI orchestration framework for building production-ready LLM applications in Python: modular pipelines and agent workflows with explicit control over retrieval, routing, memory, and generation.
An open-source AI orchestration framework for building production-ready LLM applications in Python: modular pipelines and agent workflows with explicit control over retrieval, routing, memory, and generation.Is it a framework I build on in my own code, or a ready-made app I install?It is a Python framework you code against — you compose pipelines and agents from built-in or custom components rather than install a finished product. To serve your pipelines as REST APIs or MCP servers, the companion project Hayhooks wraps them for you.
It is a Python framework you code against — you compose pipelines and agents from built-in or custom components rather than install a finished product. To serve your pipelines as REST APIs or MCP servers, the companion project Hayhooks wraps them for you.How do I get it running?Install it via pip; nightly pre-releases and Docker images are also supported. Then work through the Get Started Guide — it should take minutes to build your first LLM app, with tutorials and a Cookbook for deeper use cases.
Install it via pip; nightly pre-releases and Docker images are also supported. Then work through the Get Started Guide — it should take minutes to build your first LLM app, with tutorials and a Cookbook for deeper use cases.Do I have to commit to one LLM provider to use it?No, it is model- and vendor-agnostic: plug in OpenAI, Mistral, Anthropic, Cohere, Hugging Face, Google, Azure OpenAI, AWS Bedrock, or local models, and swap them without rewriting your system.
No, it is model- and vendor-agnostic: plug in OpenAI, Mistral, Anthropic, Cohere, Hugging Face, Google, Azure OpenAI, AWS Bedrock, or local models, and swap them without rewriting your system.Is everything free, or is there a catch?The core framework is free and open source, but expert support, enterprise-grade templates, and the managed or self-hosted Enterprise Platform are paid Haystack Enterprise offerings.
The core framework is free and open source, but expert support, enterprise-grade templates, and the managed or self-hosted Enterprise Platform are paid Haystack Enterprise offerings.Does it send any data about my usage anywhere?Yes, it collects anonymous telemetry reporting which pipeline components you initialize, and you can opt out via an option described in the docs.
Yes, it collects anonymous telemetry reporting which pipeline components you initialize, and you can opt out via an option described in the docs.What kinds of applications can I build with it?Scalable RAG systems, multimodal applications, semantic search, question answering, and autonomous agents. Pipelines support loops, branches, and conditional logic, so how context is retrieved, filtered, combined, and routed stays explicit and traceable.
Scalable RAG systems, multimodal applications, semantic search, question answering, and autonomous agents. Pipelines support loops, branches, and conditional logic, so how context is retrieved, filtered, combined, and routed stays explicit and traceable.What makes its agents production-ready?Lifecycle hooks (before_llm, before_tool, on_exit) let you add guardrails and custom logic, while step_count, token_usage, and tool calls are tracked out of the box for monitoring and cost control. Pipelines run sync or async and stream token by token, agents can make concurrent tool calls, and ready-made agents ship in Agent Pack, such as a deep research agent and an advanced RAG agent.
Lifecycle hooks (before_llm, before_tool, on_exit) let you add guardrails and custom logic, while step_count, token_usage, and tool calls are tracked out of the box for monitoring and cost control. Pipelines run sync or async and stream token by token, agents can make concurrent tool calls, and ready-made agents ship in Agent Pack, such as a deep research agent and an advanced RAG agent.Can I expose my pipelines as REST APIs or MCP servers?Yes — Hayhooks wraps pipelines and agents with custom logic and exposes them through HTTP endpoints or MCP. It also supports OpenAI-compatible chat completion endpoints and works with chat UIs like open-webui.
Yes — Hayhooks wraps pipelines and agents with custom logic and exposes them through HTTP endpoints or MCP. It also supports OpenAI-compatible chat completion endpoints and works with chat UIs like open-webui.Is it actually used in production, or mostly experimental?Thousands of teams run it in production, including Apple, Meta, Databricks, NVIDIA, Netflix, Airbus, and the European Commission. The project has been around since 2019 and is currently at version 3.0.
Thousands of teams run it in production, including Apple, Meta, Databricks, NVIDIA, Netflix, Airbus, and the European Commission. The project has been around since 2019 and is currently at version 3.0.What are similar projects?txtai, PocketFlow, adk-python, agentic-rag-for-dummies, turnstone, hive, ruflo, PageIndex
txtai, PocketFlow, adk-python, agentic-rag-for-dummies, turnstone, hive, ruflo, PageIndex