HsPy
Start Here: Haystack Reading Paths
Which Haystack pages to read, in what order, depending on what you are building.
This section covers Haystack 3 (haystack-ai 3.5.0, checked 2026-11-19). Each path below lists pages in reading order.
I’m new to Haystack
Section titled “I’m new to Haystack”- Comprehensive guide → Installation and Setup
- Architecture overview and the diagrams
- Components and Pipelines, including the minimal RAG pipeline
- Creating Agents with the Agent Class
- One recipe from the recipes page
I’m upgrading from Haystack 2.x
Section titled “I’m upgrading from Haystack 2.x”- Migrating from Haystack 2.x
- The upstream MIGRATION.md
- Observability — tracing is no longer enabled automatically
I need knowledge-base Q&A (RAG)
Section titled “I need knowledge-base Q&A (RAG)”- A Minimal RAG Pipeline
- Recipe: knowledge-base Q&A
- LLMRanker for re-ranking
I want multi-agent systems
Section titled “I want multi-agent systems”- Multi-Agent Pipeline Design Patterns
- Multi-agent guide
- Advanced agents for hooks and shared state
I need to deploy to production
Section titled “I need to deploy to production”- Production guide, including security and testing and evaluation
- Streaming server (FastAPI)
- Observability
Page guide
Section titled “Page guide”| Page | Use it for |
|---|---|
| Comprehensive guide | Core concepts, the Agent, tools, multi-agent basics, migration |
| Advanced agents | Token budgets, hooks, state handlers, custom components, async |
| Multi-agent guide | Coordinator, pipeline and routing patterns |
| Recipes | End-to-end examples to adapt |
| Diagrams | Pipelines, the agent loop and hooks, wrapping, multi-agent, spans |
| Observability | Tracing, logging, retries, fallbacks, dead letters |
| Production guide | Serving (FastAPI, Hayhooks), security, cost, testing and evaluation, scaling |
| Streaming server | Server-sent events from an Agent |
Code examples assume OPENAI_API_KEY is set. To try them without an API key, swap OpenAIChatGenerator for MockChatGenerator from haystack.components.generators.chat, which returns canned or echoed replies.