AI Agent Platforms / All platforms / Haystack
Haystack
Python orchestration framework for retrieval pipelines, document processing, and agents.
Haystack is an open-source AI orchestration framework for building production LLM applications in Python. Teams use it to build modular pipelines for retrieval, RAG, question answering, and agents with explicit control over each component.
| Category | Agent builders and frameworks |
| License | Apache-2.0 |
| Hosting | library for self-managed applications; vendor cloud available |
| Repository | github.com/deepset-ai/haystack |
| Website | haystack.deepset.ai |
| Last verified | 2026-09-02 |
Documented features
- Composes modular pipelines with explicit control over retrieval, routing, memory, and generation
- Supports conditional logic, loops, and branches in pipeline flow
- Runs synchronously or asynchronously with token-by-token streaming
- Builds agents with lifecycle hooks such as before_llm, before_tool, and on_exit
- Tracks step counts, token usage, and tool calls in agent runs
- Executes agent tools concurrently and offers ready-made templates via Agent Pack
- Supports models from OpenAI, Mistral, Anthropic, Cohere, Hugging Face, Google, Azure, AWS Bedrock, and local backends
- Exposes pipelines as REST APIs or MCP servers through Hayhooks
- Provides OpenAI-compatible chat completion endpoints
Feature summaries come from the project's own repository and documentation and describe the software as documented on the verification date, not an evaluation.