Haystack

Production-first agentic NLP. Haystack 2.x ships a component/pipeline architecture with built-in retrieval, ranking, and LLM nodes, plus an Agent class with tool calling, memory, and multi-agent orchestration patterns.

Install

pip install haystack-ai==2.28.0

Version

v2.28.0 · April 2026 · Python 3.10+

Best for

RAG pipelines, search-first agents, production deployments.

Where to start

Already know Haystack 2.x? Jump to Agent concepts or Multi-agent patterns. New? Follow Zero → Hero.

Zero → Hero

1 · Installation & setup Install, first pipeline · 10 min

2 · Architecture overview Components, pipelines, connections · 15 min

3 · Components & pipelines How building blocks compose · 20 min

4 · Agent concepts Agent loop, reasoning, state · 20 min

5 · Creating agents Agent class, configuration · 20 min

6 · Tools & function calling Built-in + custom tools · 20 min

7 · Single-purpose agents Research, summarisation, Q&A · 20 min

8 · Conversational agents Multi-turn, memory management · 15 min

9 · Agentic workflows Production patterns, resilience · 25 min

10 · Multi-agent pipelines Orchestration, delegation · 25 min

11 · Advanced agents Complex patterns, custom components · 30 min

12 · Production Deployment, scaling, monitoring · 30 min

Jump to a topic

Agent class

Tool integration

Conversational memory

Multi-agent pipelines

Error handling & resilience

SearchableToolset (v2.21+)

LLMRanker (v2.22+)

ToolInvoker

Provider-agnostic design

Observability

Reference

Comprehensive guide Single-page technical reference.

Advanced agents

Multi-agent guide 2025 multi-agent patterns.

Recipes

Diagrams

Observability

Production guide

Streaming server (FastAPI)

Start here

What’s new in v2.28.0 (April 2026)

• Stabilisation and bug fixes following 2.28 RCs.
• SearchableToolset, LLMRanker, and ToolInvoker stabilised.
• Expanded provider support (Anthropic, Gemini, Mistral, Ollama).
• Multi-agent guide published (October 2025).

Install, minimal example

from haystack.components.agents import Agent
from haystack.components.generators.chat import OpenAIChatGenerator
from haystack.dataclasses import ChatMessage

agent = Agent(
    chat_generator=OpenAIChatGenerator(model="gpt-4o"),
    system_prompt="You are a helpful research assistant.",
)

result = agent.run(messages=[ChatMessage.from_user("Summarise Haystack 2.x.")])
print(result["messages"][-1].text)

Ready? Start Zero → Hero Step 1 →

---

Revision history

Date	Version	Changes
2026-04-21	2.28.0	Index redesigned with Zero → Hero + Jump-to-topic grid.
April 20, 2026	2.28.0	GA after 2.28.0rc1/rc2; stabilisation.
November 2025	2.20.0	Initial documented version.