haystack-experimental 0.19.0.post1

Experimental components and features for the Haystack LLM framework.

Haystack experimental package

[!WARNING] This project is archived and no longer maintained.

0.19.0.post1 is the final release. There will be no further releases, bug fixes, or compatibility updates, and the repository is read-only.

• Most experiments graduated into Haystack itself. If you are looking for Agent, Tool and tool calling, AsyncPipeline, pipeline breakpoints, multimodality, SuperComponent, QueryExpander, Human-in-the-Loop and more, they all ship in haystack-ai now — see Graduated experiments below for the full list and use the Haystack documentation.

• The remaining experiments were discontinued rather than graduated — see Discontinued experiments. They are not part of Haystack and will not be maintained anywhere.

• If you depend on a discontinued experiment, pin the final release explicitly. It will remain installable from PyPI, but it is only tested against the version of Haystack that was current in February 2026 and will drift out of compatibility with newer haystack-ai releases:

  pip install "haystack-experimental==0.19.0.post1"
  

The haystack-experimental package gave Haystack users early access to experimental features without immediately committing to their official release, so that we could gather feedback and iterate quickly. Each experiment had a limited lifespan, after which it was either merged into Haystack core, released as a Core Integration, or dropped.

That process has now concluded. This README is kept as the record of where each experiment ended up.

Experiments catalog

Graduated experiments

These experiments were adopted into Haystack core and are available in haystack-ai. The version column is the last release of haystack-experimental that contained the experimental copy.

Name	Type	Final release
ChatMessage refactoring; Tool class; tool support in ChatGenerators; ToolInvoker	Tool Calling support	0.4.0
AsyncPipeline; Pipeline bug fixes and refactoring	AsyncPipeline execution	0.7.0
LLMMetadataExtractor	Metadata extraction with LLM	0.7.0
Auto-Merging Retriever & HierarchicalDocumentSplitter	Document Splitting & Retrieval Technique	0.8.0
Agent	Simplify Agent development	0.8.0
SuperComponent	Simplify Pipeline development	0.8.0
Pipeline	Pipeline breakpoints for debugging	0.12.0
ImageContent; Image Converters; multimodal support in OpenAIChatGenerator and AmazonBedrockChatGenerator; ChatPromptBuilder refactoring; SentenceTransformersDocumentImageEmbedder; LLMDocumentContentExtractor; new Routers	Multimodality	0.12.0
QueryExpander	Query Expansion Component	0.14.3
MultiQueryEmbeddingRetriever	MultiQueryEmbeddingRetriever	0.14.3
MultiQueryTextRetriever	MultiQueryTextRetriever	0.14.3
EmbeddingBasedDocumentSplitter	Document Splitting	0.15.2
Confirmation Policies; ConfirmationUIs; BlockingConfirmationStrategy; ConfirmationUIResult; ToolExecutionDecision	Human in the Loop	0.16.0
Mem0MemoryStore	MemoryStore	0.19.0

Discontinued experiments

These experiments were not adopted into Haystack. They exist only in the release listed below.

Name	Type	Final release	Cookbook	Discussion
OpenAIFunctionCaller	Function Calling Component	0.3.0	None	--
OpenAPITool	OpenAPITool component	0.3.0	Notebook	Discuss
EvaluationHarness	Evaluation orchestrator	0.7.0	None	Discuss
Agent; BreakpointConfirmationStrategy; HITLBreakpointException	Human in the Loop via Breakpoints	0.19.0	None	Discuss
InMemoryChatMessageStore; ChatMessageRetriever; ChatMessageWriter	Chat Message Store, Retriever, Writer	0.19.0		Discuss
OpenAIChatGenerator (hallucination risk scoring)	Chat Generator Component	0.19.0		Discuss
MarkdownHeaderLevelInferrer	Preprocessor	0.19.0	None	Discuss
LLMSummarizer	Document Summarizer	0.19.0	None	Discuss

Usage

Experimental features were imported like any other Haystack integration package:

from haystack import Document
from haystack.components.generators.chat import OpenAIChatGenerator
from haystack_experimental.components.summarizers import LLMSummarizer

summarizer = LLMSummarizer(chat_generator=OpenAIChatGenerator())
summarizer.run(documents=[Document(content="...")])

Some experiments came with example notebooks in the Haystack Cookbook.

Telemetry

As with the Haystack core package, this package collected anonymous usage statistics. For more information on what was collected and how to opt out, refer to the telemetry documentation.