deepset-ai/haystack
71.3
Strong · 16 November 2026
51.3k
lines of production code
Python
primary language
5
measurements over time
What this system is
Features
Add ClusterFuzzLite fuzzing infrastructure
Added configuration files for ClusterFuzzLite to enable automated fuzzing of Haystack's Atheris targets. This includes a Dockerfile pinning the \oss-fuzz-base/base-builder-python\ image, a build script that compiles fuzz harnesses with NumPy submodule support and ships seed corpora, and a project configuration specifying Python language and address sanitizers.
.clusterfuzzlite · high confidence
Add copy button to documentation pages
Documentation pages now include a sticky copy button (CopyDropdown) that allows users to easily copy content. This is implemented by wrapping the original DocItem Content component to inject the button, while the Footer component is also wrapped to maintain theme consistency.
docs-website/src/theme/DocItem · high confidence
Add copy button with AI assistant integration for documentation pages
A new CopyDropdown component has been added to the documentation website, providing users with a button to copy the current page's content as Markdown. The component uses the Turndown library to convert HTML to Markdown, preserving code blocks, inline code, and admonitions while stripping navigation and footer elements. Users can copy the raw Markdown to their clipboard or select from a dropdown menu to open the content in external AI assistants (ChatGPT, Claude, or Perplexity) with the page text pre-filled as a prompt.
docs-website/src/components/CopyDropdown · high confidence
Add platform availability badge to component documentation
A new PlatformBadge component has been added to the documentation site to visually indicate when a component is available on the Haystack Enterprise Platform. This badge, styled with a checkmark and link to the platform components page, renders automatically on doc pages that include the \hep\_available\ frontmatter field, helping users quickly identify platform-supported features.
docs-website/src/components/PlatformBadge · high confidence
Add versioned API reference documentation for v2.18
The v2.18 documentation site now includes a dedicated API reference index page that organizes technical documentation into three sections: the core Haystack framework API, official integration APIs, and experimental features.
_docs-website/reference\_versioned\docs/version-2.18 · high confidence
Added documentation templates for components and document stores
New MDX templates have been added to the documentation site to standardize how component and document store integrations are documented. The component template provides a structured layout including a key-value table for pipeline metadata (such as mandatory variables and output types) and sections for overview and usage examples. The document store template outlines sections for description, initialization, supported retrievers, and GitHub links, ensuring consistent documentation for Haystack 2.21 integrations.
(repo-wide) · high confidence
Added version 2.18 documentation for core concepts and components
The documentation site now includes the full versioned documentation set for Haystack 2.18. This update adds new concept pages covering Agents (including tool-calling and pipeline-based agents), State management for shared context, Components (stand-alone and custom), SuperComponents for wrapping pipelines, Data Classes (Answer, Document, ChatMessage, etc.), Device Management for local inference, and the general Concepts Overview. These pages provide users with the reference material needed to build and understand Haystack 2.18 pipelines.
_docs-website/versioned\docs/version-2.18 · high confidence
Automated documentation generation and release note formatting tools
Added two new scripts to the repository: \generate\_platform\_components\_table.py\ scans the Haystack codebase and integrations to automatically generate an MDX table of enterprise platform components, and \release\_note\_backticks.py\ acts as a pre-commit hook to enforce double backtick syntax for inline code in reno release notes, preventing rendering issues.
scripts · high confidence
Document experimental and stable Agents APIs for Haystack 2.26
The Haystack 2.26 reference documentation now includes dedicated API pages for the Agents component. The stable \agents\_api.md\ page documents the core \Agent\ class, covering its initialization with chat generators and tools, support for Jinja2 prompt templates, and execution logic. Additionally, the new \experiments-api/\ directory introduces reference pages for experimental features, including the \Agent\ extension with human-in-the-loop confirmation strategies, \ChatMessageStore\ and \ChatMessageRetriever\/\Writer\ components for session management, a \Mem0MemoryStore\ for external memory integration, an \OpenAIChatGenerator\ with hallucination risk scoring, a \MarkdownHeaderLevelInferrer\ for document preprocessing, and an \LLMSummarizer\ for text condensation.
_docs-website/reference\_versioned\docs/version-2.26 · high confidence
DocumentWriter now supports async execution and explicit resource closing
The DocumentWriter component in haystack.components.writers has been updated to include asynchronous execution capabilities via a new \run\_async\ method, allowing users to write documents to a DocumentStore without blocking the event loop. Additionally, \close\ and \close\_async\ methods have been added to explicitly release resources held by the underlying DocumentStore, ensuring proper cleanup of synchronous and asynchronous connections respectively.
haystack/components/writers · high confidence
Documentation for Haystack 2.22 tools and integrations
Added comprehensive documentation for the Haystack 2.22 tools ecosystem, including core abstractions like \Tool\, \Toolset\, \ComponentTool\, \PipelineTool\, and \MCPTool\/\MCPToolset\. Also added guides for the new GitHub integration tools (\GitHubFileEditorTool\, \GitHubIssueCommenterTool\, \GitHubIssueViewerTool\, \GitHubPRCreatorTool\, \GitHubRepoViewerTool\), detailing their usage in pipelines and with the Agent component.
_docs-website/versioned\docs/version-2.22/tools · high confidence
Documentation for Haystack 2.26 pipeline components
The documentation for version 2.26 has been updated to include new and existing pipeline components. This includes pages for the Agent component, which enables tool-using agents with iterative LLM interaction and state management, and Human-in-the-Loop (HITL) support, allowing users to intercept and confirm or modify agent tool calls before execution. Audio processing documentation has been added for LocalWhisperTranscriber and RemoteWhisperTranscriber. Builder components like PromptBuilder and ChatPromptBuilder are documented with details on Jinja2 template usage and variable handling. Additional components covered include CacheChecker for document store caching, and classifiers like DocumentLanguageClassifier for language detection and metadata tagging.
_docs-website/versioned\_docs/version-2.20/pipeline-components, docs-website/versioned\docs/version-2.26/pipeline-components · high confidence
Documentation for Haystack v2.30: Agents, SuperComponents, and Data Classes
This release adds comprehensive documentation for Haystack v2.30, introducing new guides for building AI agents and multi-agent systems, including how to wrap agents as tools using the \@tool\ decorator or \ComponentTool\. It also documents the \SuperComponent\ feature, which allows users to wrap entire pipelines as single reusable components, and provides detailed reference pages for core data classes like \ChatMessage\, \FileContent\, and \ImageContent\ to support multimodal interactions. Additionally, the update includes templates for component and document store documentation, and updates installation instructions to use the \weave-haystack\ package.
_docs-website/versioned\docs/version-2.30 · high confidence
Documented Haystack 2.28 experimental and core APIs
The Haystack 2.28 reference documentation now includes detailed API guides for the new experimental \haystack\_experimental\ components and the core \haystack\ components. Users can now consult the Agents API for tool-using agents with human-in-the-loop confirmation strategies, the Generators API for LLM text generation with hallucination risk scoring, and the Preprocessors API for Markdown header normalization. The documentation also covers the new Memory and ChatMessage Stores (including Mem0 and InMemory backends), Retrievers, Writers, and Summarizers within the experimental namespace, alongside the core Agents, Audio (LocalWhisperTranscriber), and other foundational components.
_docs-website/reference\_versioned\docs/version-2.28 · high confidence
Haystack 2.20 documentation site launched with core guides and migration resources
The Haystack 2.20 documentation website now includes a complete set of overview pages to help users get started and migrate from previous versions. New content covers installation (pip/conda), building a first RAG application, and a detailed migration guide from Haystack 1.x to 2.x, highlighting key changes like the package name shift from \farm-haystack\ to \haystack-ai\ and the new component-based pipeline architecture. The site also features a breaking change policy, an FAQ with links to archived 1.x and 2.17- documentation, and a comprehensive guide for migrating from LangChain/LangGraph, mapping concepts like nodes to components and graphs to pipelines.
_docs-website/versioned\docs/version-2.20/overview · high confidence
Haystack 2.25 documentation site launched with new overview pages
The Haystack 2.25 documentation website now includes a comprehensive set of overview pages to help users get started and understand the framework. New content includes a detailed breaking change policy, an installation guide supporting pip, uv, and conda, and a migration guide for moving from Haystack 1.x to 2.x. Additionally, a new guide helps users migrating from LangGraph/LangChain map their existing patterns to Haystack, and an FAQ page addresses common questions about GPU usage, telemetry, and support.
_docs-website/versioned\docs/version-2.25/overview · high confidence
Haystack 2.27 documentation site goes live with core guides and migration resources
The Haystack 2.27 documentation website has been published, providing users with essential onboarding and migration resources. New pages include a Getting Started guide with step-by-step RAG and agent examples for OpenAI, Hugging Face, and Anthropic, alongside installation instructions for pip, uv, and conda. A comprehensive migration guide details the transition from Haystack 1.x to 2.x, covering changes to package names, pipeline structures, and component mappings. Additionally, the site now features a breaking change policy, an FAQ, and an extended migration guide for users moving from LangChain/LangGraph, helping developers understand the architectural differences and map existing patterns to Haystack's ecosystem.
_docs-website/versioned\docs/version-2.27/overview · high confidence
Haystack 3.2 API reference documentation published
The versioned API reference for Haystack 3.2 is now available on the Docusaurus website. This update includes newly added documentation pages for core components such as Agents (with support for tool-using workflows, hooks, and Jinja2 user prompts), Builders (AnswerBuilder and ChatPromptBuilder), Caching (CacheChecker), and various Converters (CSV, DOCX). It also documents data classes like Breakpoint and PipelineSnapshot, document stores (including InMemoryDocumentStore with BM25 stats), document writers, embedders (Azure OpenAI support), and evaluation tools (EvaluationRunResult and evaluators like AnswerExactMatch and ContextRelevance). Additionally, the reference for the GotenbergFileConverter has been added. Also in this area: 9 commits (1 fix) modifying docs-website; 12 commits (3 fixes) modifying docs-website.
_docs-website/reference\_versioned\docs/version-3.2 · medium confidence · unverified
Haystack 3.2 documentation release
The versioned documentation for Haystack 3.2 has been published, introducing new guides for building AI agents and multi-agent systems, explaining the new SuperComponent decorator for wrapping pipelines as single components, and detailing the updated data classes (ChatMessage, FileContent, ImageContent) and custom component creation patterns.
_docs-website/versioned\docs/version-3.2 · high confidence
Haystack v3.0 documentation published
The documentation for Haystack version 3.0 is now available, introducing new concepts and components such as Agents, Multi-Agent Systems, and SuperComponents. The guide covers the updated Component API, including custom component creation and input/output mapping, and details the new data classes like ChatMessage, FileContent, and ImageContent for handling multimodal inputs. It also includes migration tips for users upgrading from previous versions.
_docs-website/versioned\docs/version-3.0 · high confidence
Initial release of the Haystack documentation website
The \docs-website\ directory is now available as a standalone Docusaurus 3 site, providing the authoritative source for Haystack documentation. This includes the main guide content, an auto-generated API reference, and a versioning system that supports multiple stable releases (2.18–3.3) alongside an unstable 'next' version. The site features a custom remark plugin to ensure cross-links between guides and API references remain version-aware, and it exposes an \llms.txt\ file for AI tool consumption. It also includes a dedicated versions page and redirects for deprecated paths.
docs-website · high confidence
Initial repository structure and contributor guidelines for Haystack v3.0
This change establishes the foundational structure for the Haystack v3.0 release, introducing key documentation and configuration files that define how the project is built, tested, and contributed to. It adds \AGENTS.md\ and \CLAUDE.md\ to provide specific guidelines for AI coding assistants, including environment setup via Hatch and API design conventions. A comprehensive \CONTRIBUTING.md\ is added, detailing the development workflow, testing requirements, and pull request standards. The \MIGRATION.md\ file documents breaking changes for users upgrading from v2.x, specifically noting that many components (such as Hugging Face, Transformers, and Whisper integrations) have been moved to external integration packages, and that tracing backends like Datadog and OpenTelemetry are now explicit connectors rather than auto-enabled dependencies. Additionally, \.gitattributes\ is configured to exclude the Docusaurus documentation site from GitHub's language statistics, and \.pre-commit-config.yaml\ sets up the initial linting and formatting hooks.
(repo-wide) · high confidence
Introduce Agent Hooks for custom run-loop interception
Users can now attach custom logic to specific points in the Agent's execution lifecycle using the new \haystack.hooks\ module. This feature provides a \Hook\ protocol and a \@hook\ decorator to wrap functions that receive the Agent's live \State\, allowing them to inspect or mutate state at defined points such as \before\_run\, \before\_llm\, \before\_tool\, \after\_tool\, \on\_exit\, and \after\_run\. The implementation includes serialization support for saving and loading hook configurations, as well as tracing integration to monitor hook invocations.
haystack/hooks · high confidence
Introduce Agent component with state management and tool tracing
This change introduces the new \Agent\ component in \haystack/components/agents\, providing a structured way to orchestrate LLM calls with tools. The \Agent\ maintains a \State\ object to track run metadata, including aggregated token usage and per-tool call counts. It supports explicit configuration for how tools read from and write to the state via \inputs\_from\_state\ and \outputs\_to\_state\ mappings. Additionally, the component implements detailed tracing for tool invocations, creating a dedicated span for each tool call to improve observability.
haystack/components/agents · high confidence
Introduce EvaluationRunResult for evaluation reporting
The \haystack/evaluation\ module now exposes the \EvaluationRunResult\ class, which stores inputs and results from evaluation pipelines and provides methods to generate aggregated or detailed reports in JSON, CSV, or DataFrame formats. This class also supports comparative reporting between two evaluation runs, allowing users to inspect and compare detailed scores across different runs.
haystack/evaluation · high confidence
Introduce Haystack 2.29 documentation with experimental agent and memory components
The v2.29 reference documentation is now available, introducing a new \agents-api\ section that documents the \Agent\ component for tool-using, provider-agnostic chat workflows. It also adds an \experiments-api\ section covering experimental features, including a tool-using \Agent\ with human-in-the-loop confirmation strategies, an \InMemoryChatMessageStore\ for session isolation, a \Mem0MemoryStore\ for external memory backends, and an \OpenAIChatGenerator\ with hallucination risk scoring. Additional experimental components include a \MarkdownHeaderLevelInferrer\ for normalizing document structure, a \ChatMessageRetriever\ and \ChatMessageWriter\ for managing conversation history, and an \LLMSummarizer\ for recursive text summarization.
_docs-website/reference\_versioned\docs/version-2.29 · high confidence
Introduce JsonSchemaValidator component for validating LLM JSON output
A new \JsonSchemaValidator\ component is added to \haystack.components.validators\, allowing users to validate the JSON content of \ChatMessage\ instances against a specified JSON Schema. The component routes messages to a \validated\ output if they conform to the schema, or to a \validation\_error\ output with a detailed error message if they do not, facilitating recovery loops with LLMs. It supports custom error templates and handles specific schema formats like OpenAI function calling schemas.
haystack/components/validators · high confidence
Introduce LinkContentFetcher for fetching web content
A new LinkContentFetcher component is available in haystack.components.fetchers, allowing users to fetch and extract content from URLs. It replaces the previous requests-based implementation with httpx, adding support for async operations and optional HTTP/2. The component handles various content types (text, binary, JSON, images, etc.), supports custom request headers, user-agent rotation on retries, and configurable timeouts and retry attempts.
haystack/components/fetchers · high confidence
Introduce Python code splitter and DOCX converter with enhanced table and link handling
Users can now split Python source code into syntax-aware chunks using the new \PythonCodeSplitter\ component, which preserves structure and handles oversized functions via secondary line-based splitting. Additionally, the \DOCXToDocument\ converter now supports configurable table formats (Markdown or CSV) and link formats, while correctly preserving hyperlinks in table cells and extracting text from text boxes.
haystack-ai · high confidence
Introduce QueryExpander component for query expansion
A new \QueryExpander\ component is now available in \haystack.components.query\. It uses a chat generator to expand a single input query into multiple semantically similar queries, which helps improve retrieval recall in RAG pipelines. The component defaults to using \OpenAIChatGenerator\ with the \gpt-4.1-mini\ model and a structured JSON prompt, but allows users to provide a custom \ChatGenerator\ or a custom \prompt\_template\. It supports configuring the number of expansions (\n\_expansions\) and whether to include the original query in the output.
haystack/components/query · high confidence
Introduce SkillStore protocol and FileSystemSkillStore implementation
Haystack now provides a pluggable skill storage layer, starting with a \FileSystemSkillStore\ that reads skills from a local directory structure. Users can configure the skills directory via a path or an environment variable (using \Secret\), and the store lazily scans for \SKILL.md\ files to build a catalog of available skills. The implementation supports reading skill instruction bodies and bundled files, including returning images as \ImageContent\ and PDFs as \FileContent\ for multimodal agents. A \SkillStore\ protocol is also introduced to allow future implementations backed by other storage systems.
_haystack/skill\stores · high confidence
Introduce SkillToolset for agent-driven skill discovery and file access
A new SkillToolset is available to enable Haystack Agents to discover and interact with skills via progressive disclosure. It exposes two fixed tools: load\_skill, which returns a skill's full instructions and a manifest of bundled files after discovering available skills during warm-up, and read\_skill\_file, which retrieves specific bundled files (including images and PDFs wrapped in ImageContent/FileContent) for multimodal consumption. The toolset is backed by a SkillStore implementation and does not support adding arbitrary tools.
haystack/tools/skills · high confidence
Introduce State class for Agent context management
The \State\ class has been added to \haystack.components.agents.state\ to provide a structured container for shared information during Agent execution. It enforces a schema where each key defines an expected type and a merge handler (defaulting to list concatenation for lists and value replacement for other types). The class includes built-in serialization support via \to\_dict\ and \from\_dict\, automatically manages a \messages\ field of type \list\[ChatMessage\]\, and allows skipping specific keys during serialization.
haystack/components/agents/state · high confidence
Introduce SuperComponent to wrap Pipelines as single components
Users can now wrap an entire Haystack Pipeline inside a SuperComponent, exposing the pipeline's inputs and outputs as a single component interface. This allows pipelines to be nested within other pipelines or used directly with agents. The SuperComponent supports custom input and output mappings to rename or select specific pipeline sockets, handles lifecycle management (warm-up, close) for the wrapped pipeline, and provides both synchronous and asynchronous execution paths. It also includes type compatibility checks to ensure mapped inputs and outputs align correctly.
_haystack/core/super\component · high confidence
Introduce YAML-based component serialization with tuple support
The \haystack/marshal\ module has been introduced to handle component serialization, providing a \Marshaller\ protocol and a \YamlMarshaller\ implementation. This change enables pipelines to be serialized to and loaded from YAML format. A key behavioral addition is the support for Python tuples in YAML serialization and deserialization, allowing tuple data structures to be preserved during the marshal/unmarshal cycle. The implementation uses custom YAML loaders and dumpers to handle these types safely.
haystack/marshal · high confidence
Introduce context compaction for Agents to manage long conversations
Agents can now automatically compress their conversation history to stay within the model's context window. This is achieved by registering a \CompactionHook\ on the Agent, which triggers a \Compactor\ when the conversation size exceeds a configured threshold. The module provides three compaction strategies: \SlidingWindowCompactor\ (drops oldest messages while preserving tool-call integrity), \SummarizationCompactor\ (uses an LLM to summarize older turns), and \ToolResultPruningCompactor\ (replaces large tool outputs with placeholders). This allows Agents to maintain longer interactions without hitting context limits, with the trade-off being lossy history retention.
haystack/hooks/compaction · high confidence
Introduce new Haystack v3 converter components and OutputAdapter
The \haystack/components/converters\ package has been restructured for Haystack v3, introducing a suite of new file conversion components: \CSVToDocument\ (with row-level conversion mode), \DOCXToDocument\ (with table and hyperlink formatting options), \FileToFileContent\ (for generating base64-encoded file contents), \HTMLToDocument\ (using the Trafilatura backend), \JSONConverter\ (with jq filtering), \MarkdownToDocument\ (with YAML frontmatter extraction), \MSGToDocument\ (extracting email metadata and attachments), \PDFMinerToDocument\, \PPTXToDocument\, \PyPDFToDocument\, \TextFileToDocument\, and \XLSXToDocument\ (with hyperlink support). Additionally, the \MultiFileConverter\ super-component is provided to route and convert multiple file types in a single pipeline, and the \OutputAdapter\ component allows adapting component outputs using Jinja templates.
haystack/components/converters · high confidence
Introduce new PromptBuilder, ChatPromptBuilder, and AnswerBuilder components
The \haystack.components.builders\ package now exposes three new components: \PromptBuilder\ and \ChatPromptBuilder\ for rendering Jinja2 templates (with \ChatPromptBuilder\ supporting both string and \ChatMessage\ list templates, optional variables, and a sandboxed Jinja2 environment), and \AnswerBuilder\ for converting generator replies into \GeneratedAnswer\ objects with configurable regex patterns, document reference parsing, and optional reference range expansion. These components replace the legacy \DynamicPromptBuilder\ and \DynamicChatPromptBuilder\ classes, which have been removed.
haystack/components/builders · high confidence
Introduce new chat generator components and lazy-loading module structure
The \haystack/components/generators/chat\ package is restructured to use lazy imports via \LazyImporter\, exposing \OpenAIChatGenerator\, \OpenAIResponsesChatGenerator\, \AzureOpenAIChatGenerator\, \AzureOpenAIResponsesChatGenerator\, \FallbackChatGenerator\, \LLM\, and \MockChatGenerator\. This change adds \MockChatGenerator\ for deterministic testing without external API calls, introduces \FallbackChatGenerator\ to sequentially try multiple generators until one succeeds, and provides \LLM\ as a simplified, tool-free wrapper around \Agent\ for direct text generation. The Azure generators are updated to support the new Responses API alongside the standard Chat Completion API, and all components now support async execution and tool-based workflows.
haystack/components/generators/chat · high confidence
Introduce new core pipeline package with base orchestration, security, and visualization
The \haystack/core/pipeline\ package has been restructured into a new set of modules that define the pipeline execution engine and its supporting infrastructure. \base.py\ introduces \PipelineBase\, which manages the component graph, execution ordering, and serialization, while \pipeline.py\ provides the concrete \Pipeline\ class with synchronous and asynchronous run capabilities, including streaming and breakpoint support. \breakpoint.py\ adds logic for saving and loading pipeline snapshots, controlled by the \HAYSTACK\_PIPELINE\_SNAPSHOT\_SAVE\_ENABLED\ environment variable. \component\checks.py\ isolates the logic for determining component readiness and trigger conditions. \draw.py\ implements pipeline visualization using Mermaid, including validation of server responses for security. \descriptions.py\ provides utilities to identify pipeline inputs and outputs, and \utils.py\ contains helper functions like a safe deep-copy mechanism and a FIFO priority queue for execution scheduling. Additionally, \\\init\\_.py\ exposes the main \Pipeline\ class.
haystack/core/pipeline · high confidence
Introduce new document preprocessing components and a unified preprocessor
The \haystack/components/preprocessors\ module now includes several new components for document cleaning and splitting: \CSVDocumentCleaner\ removes empty rows and columns from CSV content; \CSVDocumentSplitter\ splits CSV documents by row thresholds or row-wise; \TextCleaner\ provides text normalization (Unicode, ASCII) and cleaning; \MarkdownHeaderSplitter\ splits documents by Markdown headers; \PythonCodeSplitter\ splits Python code; \RecursiveDocumentSplitter\ handles recursive chunking; and \EmbeddingBasedDocumentSplitter\ splits documents based on embedding similarity. Additionally, \DocumentPreprocessor\ is introduced as a SuperComponent that chains \DocumentSplitter\ and \DocumentCleaner\ together, simplifying the common preprocessing workflow.
haystack/components/preprocessors · high confidence
Introduce new evaluation components for retrieval and LLM-based metrics
This change adds a suite of new evaluator components to the \haystack.components.evaluators\ package. For retrieval evaluation, it introduces \DocumentMAPEvaluator\, \DocumentMRREvaluator\, \DocumentNDCGEvaluator\, and \DocumentRecallEvaluator\, which calculate Mean Average Precision, Mean Reciprocal Rank, Normalized Discounted Cumulative Gain, and Recall scores respectively, all supporting configurable document comparison fields. For LLM-based evaluation, it adds \LLMEvaluator\ as a base class and specific implementations: \FaithfulnessEvaluator\ to check if answers are grounded in context, \ContextRelevanceEvaluator\ to assess context relevance to questions, \SASEvaluator\ for semantic answer similarity, and \AnswerExactMatchEvaluator\ for exact string matching. The \\_\init\\_.py\ file exports all these new classes for easy import.
haystack/components/evaluators · high confidence
Introduce new tools package with lazy imports and lifecycle management
The \haystack/tools\ package has been reorganized to support lazy loading of tool components, significantly reducing import times. This change introduces a new \\_\init\\_.py\ that uses \LazyImporter\ to defer loading of tools like \ComponentTool\, \PipelineTool\, and \SearchableToolset\ until they are actually used, while eagerly importing core decorators like \@tool\ to avoid naming collisions. Additionally, the package now includes dedicated utility functions (\warm\_up\_tools\, \close\_tools\, etc.) and a \ToolsType\ alias to standardize how tools and toolsets are initialized, managed, and serialized across the Haystack ecosystem.
haystack/tools · high confidence
Introduce token counting components for chat conversations
The \haystack.token\_counters\ module now provides three implementations for estimating or determining the number of tokens in a list of \ChatMessage\s: \ApproximateTokenCounter\ (a local character-to-token ratio estimator), \TiktokenCounter\ (a local count using OpenAI's \tiktoken\ encoder), and \OpenAITokenCounter\ (which queries OpenAI's input tokens API for an exact count). All counters support counting tool schemas alongside messages and handle non-text content like images and files via configurable flat estimates. A \TokenCounter\ protocol defines the interface, and utility functions handle rendering conversations and tool specs for measurement.
_haystack/token\counters · high confidence
Introduce tool result offloading to reduce agent context usage
Agents can now offload large tool results to an external store, replacing them in the conversation with compact pointers to save context window space. This is controlled by registering a \ToolResultOffloadHook\ with an \Agent\, which uses configurable \OffloadPolicy\ strategies (such as \AlwaysOffload\, \NeverOffload\, or \OffloadOverChars\) to decide which tools' results to offload. The system includes a \FileSystemToolResultStore\ for local persistence and supports binary content like images and files. If a store write fails or a tool produces an error, the result remains in context to ensure the agent continues running without interruption.
_haystack/hooks/tool\_result\offloading · high confidence
Introduction of ChatGenerator protocol
A new \ChatGenerator\ protocol has been added to define the minimal interface for chat generator components. This protocol specifies that implementations must provide a \run\ method accepting a list of \ChatMessage\ objects and returning a dictionary, allowing for flexible parameter handling in subclasses.
haystack/components/generators/chat/types · high confidence
New API endpoints for search and MCP proxying on the documentation site
The documentation website now exposes serverless API endpoints to power interactive features. A new \/api/search\ endpoint allows the frontend to query the Deepset search API, supporting optional filtering by document type. Additionally, a \/api/mcp\ endpoint acts as a proxy for the Model Context Protocol, forwarding JSON-RPC requests to the Deepset MCP service while handling CORS and session IDs. A \/api/well-known\ endpoint returns a JSON 404 to satisfy OAuth discovery probes required by MCP clients, and a \tsconfig.json\ has been added to configure TypeScript compilation for these new API files.
docs-website/api · high confidence
New CacheChecker component for document store lookups
A new \CacheChecker\ component is introduced in \haystack.components.caching\ to check for the presence of documents in a Document Store based on a specified metadata field. It returns matching documents as 'hits' and non-matching items as 'misses'. The component supports both synchronous and asynchronous execution via \run\ and \run\_async\ methods, and includes \close\ and \close\_async\ methods to properly release resources held by the underlying Document Store.
haystack/components/caching · high confidence
New ClickableImage component with zoom functionality
The documentation site now includes a new ClickableImage component that allows users to click on images to view them in a larger, zoomed overlay. This component supports standard and large sizes, uses Docusaurus's IdealImage for rendering, and includes accessibility features like aria-pressed and role attributes. The implementation provides a better user experience for viewing detailed documentation images.
docs-website/src/components/ClickableImage · high confidence
New Human-in-the-Loop confirmation hook for Agents
The Human-in-the-Loop (HITL) module has been moved under the Hooks package and restructured to provide a \ConfirmationHook\ for Agents. This hook intercepts tool calls before execution, allowing users to confirm, reject, or modify tool parameters via configurable strategies. The module introduces \BlockingConfirmationStrategy\ which integrates with \ConfirmationPolicy\ implementations (such as \AlwaysAskPolicy\, \NeverAskPolicy\, and \AskOncePolicy\) and \ConfirmationUI\ interfaces (including \RichConsoleUI\ and \SimpleConsoleUI\) to manage user interaction. Data classes \ConfirmationUIResult\ and \ToolExecutionDecision\ define the structure for user feedback and execution outcomes, enabling precise control over tool execution flow within the Agent lifecycle.
_haystack/hooks/human\_in\_the\loop · high confidence
New LLM and Regex text extraction components
The \haystack.components.extractors\ package now exposes two new components: \LLMMetadataExtractor\, which uses a \ChatGenerator\ to extract structured metadata from documents based on a Jinja2 prompt, and \RegexTextExtractor\, which captures text from strings or chat messages using a regular expression pattern. Both components are lazily imported via the package's \\_\init\\_.py\.
haystack/components/extractors · high confidence
New LLMDocumentContentExtractor for image-based documents
A new \LLMDocumentContentExtractor\ component is introduced in \haystack.components.extractors.image\ to extract textual content and metadata from image-based documents (such as scanned PDFs or images) using a vision-enabled LLM. The component converts documents to images via \DocumentToImageContent\, sends them to a provided \ChatGenerator\, and parses the LLM response to populate document content and metadata. It supports configurable prompts, parallel processing via \max\_workers\, and handles failures gracefully by returning failed documents in a dedicated field or raising exceptions based on the \raise\_on\_failure\ setting. Security is addressed by allowing a \root\_path\ to restrict file access to a specific directory, preventing path traversal attacks.
haystack/components/extractors/image · high confidence
New LLMRanker and LostInTheMiddleRanker components added to the rankers module
The \haystack/components/rankers\ module now includes two new ranking components alongside the existing \MetaFieldRanker\ and \MetaFieldGroupingRanker\. The \LLMRanker\ allows users to rerank documents using a Large Language Model (defaulting to OpenAI's \gpt-4.1-mini\ via a \ChatGenerator\) by sending a prompt that instructs the model to return ranked document indices as JSON; it supports custom prompts, \top\_k\ limits, and fallback behavior on generation failures. The \LostInTheMiddleRanker\ reorders documents based on the 'lost in the middle' phenomenon, placing the most relevant documents at the beginning and end of the context while pushing less relevant ones to the middle, which is useful for preparing context for LLMs; it also supports \top\_k\ and \word\_count\threshold\ parameters and deduplicates documents by ID before reordering. Both new components are exported via the \\\init\\_.py\ module and follow the standard Haystack component interface.
haystack/components/rankers · high confidence
New OpenAI Decisions-based document classifier component
Users can now classify documents using OpenAI's Decisions API via the new \OpenAIDecisionsDocumentClassifier\. This component supports three question types—choice, score, and predicate—to extract structured metadata from document content, with answers stored in a configurable metadata field. It handles asynchronous execution, retries, and error reporting, allowing users to integrate OpenAI's decision models directly into their Haystack pipelines.
haystack/components/classifiers · high confidence
New OpenAIImageGenerator component and streaming utility updates
Users can now generate images using OpenAI's models (defaulting to gpt-image-2) via the new OpenAIImageGenerator component, which supports synchronous and asynchronous execution, configurable quality/size, and custom HTTP client settings. Additionally, the generators module now includes updated streaming utilities: print\_streaming\_chunk provides improved formatting for tool calls, reasoning content, and normal text, while \_convert\_streaming\_chunks\_to\_chat\_message better handles tool call accumulation, reasoning content, finish reasons, and usage info across streaming chunks.
haystack/components/generators · high confidence
New ProductOnly component and ClickableImage MDX integration
The documentation site now includes a new \ProductOnly\ component that conditionally renders content based on the current product context (derived from site config or environment variables), allowing users to view documentation specific to their selected product while keeping anchors visible for table-of-contents purposes. Additionally, the MDX theme has been extended to support a \ClickableImage\ component, enabling interactive image handling in documentation pages.
docs-website/src/components/ProductOnly, docs-website/src/theme/MDXComponents · high confidence
New TokenBudgetHook to stop Agent runs when token limits are reached
Users can now attach a \TokenBudgetHook\ to an Agent to enforce a maximum cumulative token usage. The hook monitors token consumption at the \before\_llm\ hook point and, if the budget is exceeded, stops the Agent run with the exit reason \token\_budget\_exceeded\. It optionally appends a final assistant message explaining the stop. This feature allows for better cost control and resource management in long-running Agent workflows.
haystack/hooks/budget · high confidence
New and updated documentation for document store integrations
The documentation site now includes dedicated pages for several new and existing document store integrations, providing users with installation instructions, initialization examples, and details on supported retrievers. New additions cover ArcadeDB, Astra, Azure AI Search, Chroma, Elasticsearch, FAISS (including a troubleshooting section for OpenMP conflicts on macOS), In-Memory, MongoDB Atlas, OpenSearch, Pgvector, Pinecone, Qdrant, Valkey, and Weaviate.
_docs-website/versioned\docs/version-2.25/document-stores · high confidence
New centralized utility module for authentication, serialization, and device management
The \haystack.utils\ package has been introduced as a central hub for core infrastructure utilities, exposing key components such as the \Secret\ class for structured authentication, comprehensive callable and type serialization/deserialization functions, and framework-agnostic device management classes (\ComponentDevice\, \DeviceMap\). This module also provides utilities for document filtering, async execution helpers, and secure Jinja2 extensions, consolidating previously scattered or internal logic into a public, lazily-imported interface for users building and configuring pipelines.
haystack/utils · high confidence
New core package structure with enhanced type validation and pipeline error handling
The \haystack/core\ module has been reorganized into a new package structure, introducing \haystack/core/\_\init\\_.py\ which exposes the \SuperComponent\ class. This change includes a new \haystack/core/errors.py\ module that defines specific exception classes like \PipelineRuntimeError\ and \BreakpointException\, which now carry detailed context such as component names, types, and pipeline snapshots to improve debugging. Additionally, \haystack/core/type\_utils.py\ provides robust type resolution and validation logic, supporting modern Python typing features (including PEP 604 union syntax \X\|Y\) and enabling flexible connections between types like \list\[T\]\ and \Iterable\[T\]\, as well as automatic conversion between \ChatMessage\ and \str\.
haystack/core · high confidence
New deployment and observability documentation for Haystack 2.22
Added comprehensive guides for deploying Haystack pipelines using Docker, Kubernetes, and OpenShift, including specific instructions for integrating with Hayhooks. The update also introduces new documentation for enabling GPU acceleration, configuring logging (standard, real-time, and structured), and setting up tracing backends such as OpenTelemetry, Datadog, Langfuse, and Weights & Biases.
_docs-website/versioned\docs/version-2.22/development · high confidence
New document store integrations and ComponentTool documentation added
The documentation now includes guides for several new document stores: ArcadeDB, Astra, Azure AI Search, Chroma, Elasticsearch, FAISS, InMemory, MongoDB Atlas, OpenSearch, Pgvector, Pinecone, Qdrant, Valkey, and Weaviate. Each guide covers installation, initialization, and supported retrievers. Additionally, documentation for the ComponentTool wrapper, which allows Haystack components to be used as tools by LLMs, has been added.
_docs-website/versioned\_docs/version-2.27/document-stores, docs-website/versioned\docs/version-2.27/tools · high confidence
New document store integrations for Haystack 2.20
This release adds documentation for several new document store integrations, including Astra, Azure AI Search, Chroma, Elasticsearch, In-Memory, MongoDB Atlas, OpenSearch, Pgvector, Pinecone, Qdrant, and Weaviate. Each entry provides installation instructions, initialization examples, and details on supported retrievers, enabling users to easily integrate these vector databases into their Haystack pipelines.
_docs-website/versioned\docs/version-2.20/document-stores · high confidence
New documentation for Agent, Human-in-the-Loop, and Audio components
The documentation site now includes dedicated pages for the \Agent\ component, which enables iterative, tool-using LLM workflows, and the \Human-in-the-loop\ (HITL) system, allowing users to intercept, confirm, reject, or modify agent tool calls before execution. Additionally, new documentation has been added for audio processing, covering \LocalWhisperTranscriber\ and \RemoteWhisperTranscriber\ for local and API-based speech-to-text transcription, along with their respective external integrations.
_docs-website/versioned\docs/version-2.25/pipeline-components · high confidence
New documentation for Agents, Multi-Agent Systems, and Component Templates
The documentation site now includes dedicated pages explaining how to build AI agents in Haystack, including a new guide on Multi-Agent Systems where coordinator agents delegate tasks to specialist agents. It also introduces a standardized component documentation template and a new concept page for SuperComponents, which allow users to wrap entire pipelines as single reusable components.
_docs-website/versioned\docs/version-2.28 · high confidence
New documentation for Agents, State, and Component architecture
The concepts documentation has been expanded with new pages explaining how to build AI agents in Haystack, including the \Agent\ component, tool calling, and the new \State\ container for sharing data between tools. It also introduces \SuperComponent\ for wrapping pipelines as single components, provides a comprehensive overview of core data classes like \ChatMessage\ and \Document\, and details the framework-agnostic device management system for local model inference.
_docs-website/versioned\docs/version-2.20/concepts · high confidence
New documentation for Agents, State, and SuperComponents
The documentation site now includes dedicated guides for building AI agents, managing execution state, and wrapping pipelines as components. The new 'Agents' page explains how to use the Agent component and ToolInvoker to create systems that retrieve information, generate responses, and take actions. A 'State' guide details how to use the State container to share data and messages between tools during agent execution. Additionally, 'SuperComponents' documentation shows how to wrap complete pipelines into single components using the @super\_component decorator or the SuperComponent class, simplifying complex interfaces for reuse.
_docs-website/versioned\docs/version-2.23 · high confidence
New documentation for AsyncPipeline, Breakpoints, Loops, and Templates
The documentation for the \concepts/pipelines\ section has been updated with new pages covering \AsyncPipeline\ (including concurrent execution and async generators), \Pipeline Breakpoints\ (for pausing, inspecting, and resuming execution via snapshots), \Pipeline Loops\ (feedback flows and safety limits), and \Pipeline Templates\ (predefined pipelines like RAG and Indexing). Existing pages for creating, debugging, serializing, and visualizing pipelines have also been refreshed to reflect these capabilities.
_docs-website/versioned\docs/version-2.21/concepts/pipelines · high confidence
New documentation for Haystack 2.21 core concepts
This release adds new documentation pages for Haystack 2.21, covering the \State\ container for agent execution, guidelines for creating custom components, the \SuperComponent\ wrapper for pipelines, and the \ChatMessage\ data class for LLM interactions.
_docs-website/versioned\_docs/version-2.21/concepts/agents, docs-website/versioned\_docs/version-2.21/concepts/components, docs-website/versioned\docs/version-2.21/concepts/data-classes · high confidence
New documentation for Haystack 2.22 concepts
The documentation site now includes a new 'concepts' section for Haystack 2.22, introducing pages that explain core architectural elements. This includes an overview of agents, their components (like ToolInvoker and ComponentTool), and how to build them using pipelines or the Agent class. It also introduces the State container for managing shared information and intermediate results during agent execution. Additionally, new pages cover the fundamentals of components, how to create custom components, and the SuperComponent pattern for wrapping entire pipelines as single units. The documentation also details the core data classes (Document, ChatMessage, Answer, etc.) and the framework-agnostic device management system for local model inference.
_docs-website/versioned\_docs/version-2.22/concepts, docs-website/versioned\_docs/version-2.26/concepts, docs-website/versioned\_docs/version-2.25/concepts, docs-website/versioned\docs/version-2.27/concepts · high confidence
New documentation for Haystack 2.24 concepts and templates
The documentation site now includes a comprehensive set of new pages for version 2.24, covering core concepts such as Agents, State management, Components, SuperComponents, and Data Classes. It also introduces reusable MDX templates for generating component and document store documentation, providing users with structured guides on building custom components, managing agent state, and understanding the underlying data abstractions like ChatMessage and Document.
_docs-website/versioned\docs/version-2.24 · high confidence
New documentation for Haystack 2.25 tools and integrations
The Haystack 2.25 documentation now includes dedicated pages for the new \ComponentTool\ and \PipelineTool\ wrappers, which allow users to expose existing Haystack components and pipelines as LLM tools. It also introduces documentation for \MCPTool\ and \MCPToolset\ for Model Context Protocol integrations, highlighting the deprecation of SSE transport in favor of Streamable HTTP. Additionally, new pages cover the GitHub integration tools (\GitHubFileEditorTool\, \GitHubIssueCommenterTool\, \GitHubIssueViewerTool\, \GitHubPRCreatorTool\, \GitHubRepoViewerTool\) and the \SearchableToolset\ for managing large tool catalogs.
_docs-website/versioned\docs/version-2.25/tools · high confidence
New documentation for Haystack 3.1 core concepts and component templates
This update introduces the foundational documentation structure for Haystack version 3.1, including new concept pages for Agents, Multi-Agent Systems, Components, SuperComponents, and Data Classes. It also adds reusable MDX templates for component and document store documentation to standardize future integration guides.
_docs-website/versioned\docs/version-3.1 · high confidence
New documentation for advanced RAG techniques and pipeline evaluation
The Haystack 2.20 documentation now includes dedicated guides for optimizing retrieval and evaluating pipelines. Users can learn about Advanced RAG Techniques, specifically the Hypothetical Document Embeddings (HyDE) method, with a complete code example demonstrating how to generate hypothetical documents to improve retrieval in unseen domains. Additionally, new pages cover Evaluation strategies, detailing both Model-Based evaluation (using LLMs or local models for metrics like faithfulness and context relevance) and Statistical evaluation (using metrics like recall, MRR, and exact match), including integrations with Ragas and DeepEval.
_docs-website/versioned\_docs/version-2.20/optimization, docs-website/versioned\docs/version-2.27/optimization · high confidence
New documentation for choosing and creating custom document stores
Added two new documentation pages to the Haystack 2.21 concepts guide: 'Choosing a Document Store' and 'Creating Custom Document Stores'. The former helps users select the right vector database by categorizing options (pure vector, SQL, NoSQL, full-text, in-memory) and comparing their trade-offs. The latter provides a comprehensive guide for developers building custom integrations, covering naming conventions, project structure, packaging for PyPI, implementing the DocumentStore protocol, handling secrets, and writing tests.
_docs-website/versioned\docs/version-2.21/concepts/document-store · high confidence
New documentation for deploying Haystack pipelines and configuring observability
Added comprehensive guides for deploying Haystack pipelines using Docker, Kubernetes, OpenShift, and the Hayhooks REST API server. The update also introduces new documentation pages for enabling GPU acceleration, configuring standard and structured logging, and setting up tracing backends including OpenTelemetry, Datadog, Langfuse, MLflow, and Weights & Biases Weave.
_docs-website/versioned\docs/version-2.25/development · high confidence
New documentation for deploying Haystack pipelines and using Hayhooks
Added new documentation pages covering deployment strategies for Haystack pipelines, including guides for Docker, Kubernetes, and OpenShift. The release also introduces documentation for Hayhooks, a web application for serving Haystack pipelines as REST APIs, detailing its installation, configuration, and usage for pipeline deployment and file uploads. Additionally, new pages explain how to enable GPU acceleration for Transformer models and provide an overview of external integrations for tracing and monitoring.
_docs-website/versioned\_docs/version-2.20/development, docs-website/versioned\docs/version-2.27/development · high confidence
New documentation for deployment, GPU acceleration, and observability
Added new documentation pages covering how to deploy Haystack pipelines using Docker, Kubernetes, OpenShift, and Hayhooks, as well as guides for enabling GPU acceleration and configuring logging and tracing backends like OpenTelemetry and Datadog.
_docs-website/versioned\docs/version-2.21/development · high confidence
New documentation for version 2.19 concepts
Added the initial documentation pages for version 2.19, covering core concepts such as Agents, State, Components, Custom Components, SuperComponents, Data Classes (including ChatMessage), Device Management, and Document Stores.
_docs-website/versioned\docs/version-2.19 · high confidence
New documentation pages for Haystack 2.22 overview
The versioned documentation for Haystack 2.22 now includes several new overview pages: a Breaking Change Policy detailing versioning conventions and deprecation processes, an FAQ addressing common questions about GPU usage and telemetry, a Get Started guide with examples for OpenAI, Hugging Face, and Anthropic, an Installation page with pip and conda instructions, a Migration Guide for moving from Haystack 1.x to 2.x, a LangGraph/LangChain migration guide, and a Telemetry page explaining data collection and opt-out methods.
_docs-website/versioned\_docs/version-2.21/overview, docs-website/versioned\docs/version-2.22/overview · high confidence
New embedder components and lazy-loading module structure
The \haystack/components/embedders\ package now exposes \OpenAIDocumentEmbedder\, \OpenAITextEmbedder\, \AzureOpenAIDocumentEmbedder\, \AzureOpenAITextEmbedder\, \MockDocumentEmbedder\, and \MockTextEmbedder\ via a lazy-import \\_\init\\_.py\. The OpenAI embedders default to the \text-embedding-3-small\ model and support async execution, configurable \dimensions\, \timeout\, \max\_retries\, and \raise\_on\_failure\. The Azure embedders inherit from their OpenAI counterparts, adding Azure-specific configuration (endpoint, deployment, API version, AD token) while retaining async support. The new mock embedders provide deterministic, API-free embeddings for testing and prototyping, supporting fixed vectors, custom callables, and text-hashed deterministic outputs.
haystack/components/embedders · high confidence
New experimental and stable API reference documentation for Haystack v2.25
The v2.25 documentation site now includes comprehensive reference pages for the new Agents API and a suite of Experimental components. The stable Agents API (\haystack.components.agents\) is documented with support for tool-using agents, user prompt templates with Jinja2 variables, and human-in-the-loop confirmation strategies. The Experimental API (\haystack\_experimental\) introduces documentation for new components including an Agent with human-in-the-loop capabilities, an OpenAI Chat Generator with hallucination risk scoring, a Markdown Header Level Inferrer, an LLM Summarizer, and a full set of Chat Message management components (InMemoryChatMessageStore, ChatMessageWriter, ChatMessageRetriever) alongside a Mem0-based Memory Store.
_docs-website/reference\_versioned\docs/version-2.25 · high confidence
New filter merging policy and document store protocol types
This change introduces a new \FilterPolicy\ enum with \REPLACE\ and \MERGE\ options, along with the \apply\_filter\_policy\ function that allows runtime filters to be merged with initialization filters (with runtime values overwriting init values). It also defines the \DocumentStore\ protocol and \DuplicatePolicy\ enum, establishing the type contracts for document storage operations like filtering and writing documents with conflict resolution strategies.
_haystack/document\stores/types · high confidence
New image conversion components for multimodal pipelines
Added a new \haystack.components.converters.image\ package containing four components to support multimodal workflows: \DocumentToImageContent\ converts existing Documents (referencing image or PDF files) into \ImageContent\ objects, including batched PDF page extraction; \ImageFileToDocument\ wraps image file paths into empty Documents for downstream processing; \ImageFileToImageContent\ converts image files directly into \ImageContent\ objects with optional resizing; and \PDFToImageContent\ converts PDF files into \ImageContent\ objects with configurable page ranges. All components support optional image resizing and detail levels, and \DocumentToImageContent\ includes security measures to prevent path-traversal attacks when resolving file paths.
haystack/components/converters/image · high confidence
New joiner components and package structure
The joiner components have been reorganized into a new \haystack.components.joiners\ package, exposing \AnswerJoiner\, \BranchJoiner\, \DocumentJoiner\, \ListJoiner\, and \StringJoiner\. \AnswerJoiner\ merges answer lists with optional score-based sorting and top-k filtering. \BranchJoiner\ merges multiple pipeline branches by forwarding the first available input, useful for loop handling and decision-based merging. \DocumentJoiner\ supports multiple join modes including concatenate, merge, reciprocal rank fusion, and distribution-based rank fusion, with configurable weights and top-k limits. \ListJoiner\ flattens multiple lists of the same type into a single list, while \StringJoiner\ combines strings from different components into a list.
haystack/components/joiners · high confidence
New modular routers package with updated routing components
The \haystack.components.routers\ package has been introduced to centralize routing logic, exporting \ConditionalRouter\, \DocumentLengthRouter\, \DocumentTypeRouter\, \FileTypeRouter\, \LLMMessagesRouter\, and \MetadataRouter\. \ConditionalRouter\ now supports \output\_passthrough\ for routing complex non-basic types without Jinja2 serialization, allows custom Jinja filters, and validates output types. \LLMMessagesRouter\ is a new component that uses a \ChatGenerator\ to classify chat messages based on LLM output patterns. \MetadataRouter\ now supports routing \ByteStream\ objects in addition to \Document\s and includes a \strict\_datetime\_comparison\ option for precise timezone handling. \FileTypeRouter\ and \DocumentTypeRouter\ both support regex-based MIME type matching and allow passing metadata to \ByteStream\ sources.
haystack/components/routers · high confidence
New protocols for Text and Document Embedders
A new \haystack.components.embedders.types\ module has been introduced, defining \TextEmbedder\ and \DocumentEmbedder\ protocols. These protocols standardize the interface for embedder components, specifying that \TextEmbedder.run\ accepts a string and returns a dictionary with an embedding, while \DocumentEmbedder.run\ accepts a list of documents and returns a dictionary with the embedded documents. This provides a consistent contract for implementing text and document embedding logic.
haystack/components/embedders/types · high confidence
New retriever components and lazy-loading module structure
The \haystack.components.retrievers\ package now exposes a set of new retrieval components: \AutoMergingRetriever\ (merges hierarchical leaf documents into parents based on a threshold), \FilterRetriever\ (filters documents by metadata), \MultiRetriever\ (runs multiple text retrievers in parallel with configurable join modes like reciprocal rank fusion), \MultiQueryEmbeddingRetriever\ and \MultiQueryTextRetriever\ (run multiple queries in parallel and deduplicate results), \SentenceWindowRetriever\ (fetches surrounding context documents), and \TextEmbeddingRetriever\ (wraps an embedding retriever with a text embedder). The module \\_\init\\_.py\ has been replaced with a lazy-import structure to optimize import times, making these components available via \from haystack.components.retrievers import ...\.
haystack/components/retrievers · high confidence
New scripts for local docs development and sidebar extraction
Added \setup-dev.sh\ to automate the local development environment setup for the Haystack documentation site, including dependency installation with a time-bound pip guard, and \extract\_sidebar.mjs\ to convert Docusaurus sidebar configuration files into JSON format for easier processing.
docs-website/scripts · high confidence
New unified tracing subsystem with content tracing and payload optimization
The haystack/tracing module introduces a new abstraction for distributed tracing, providing \Span\ and \Tracer\ interfaces along with a \ProxyTracer\ to manage global state. Users can now enable or disable tracing via \enable\_tracing\ and \disable\_tracing\, and control the inclusion of sensitive data (queries, documents, answers) using the \HAYSTACK\_CONTENT\_TRACING\_ENABLED\ environment variable. To prevent large payloads, tag values are automatically coerced to primitive types or serialized with placeholders for complex objects.
haystack/tracing · high confidence
Promote Haystack 2.21 concepts documentation to stable
The core concept documentation for Haystack 2.21—including guides on Agents, Components, Data Classes, Device Management, Document Stores, Integrations, Jinja Templates, Metadata Filtering, and Pipelines—has been promoted from unstable to stable status. This makes these reference pages officially available for users building and configuring Haystack pipelines.
_docs-website/versioned\docs/version-2.21/concepts · high confidence
Promote Haystack 2.21 document store and tool documentation to stable
The documentation for several document stores (Astra, Azure AI Search, Chroma, Elasticsearch, In-Memory, MongoDB Atlas, OpenSearch, Pgvector, Pinecone, Qdrant, and Weaviate) and tools (ComponentTool, MCPTool, and MCPToolset) has been promoted from unstable to stable status for version 2.21. This makes these integration guides officially available and recommended for use in this release.
_docs-website/versioned\_docs/version-2.21/document-stores, docs-website/versioned\docs/version-2.21/tools · high confidence
Promote Haystack 2.26 development documentation
The Haystack 2.26 development documentation is now stable and officially promoted, providing comprehensive guides for deploying pipelines via Docker, Kubernetes, and OpenShift, as well as using Hayhooks to serve pipelines as REST APIs. The release also includes updated instructions for enabling GPU acceleration, configuring logging and tracing with backends like OpenTelemetry and Datadog, and integrating with external tools such as Arize, Ray, and Weights & Biases.
_docs-website/versioned\docs/version-2.26/development · high confidence
Promote Haystack 2.27 documentation and publish new experimental API references
The Haystack 2.27 documentation has been promoted to stable status, making the latest stable release docs publicly available. Additionally, a comprehensive set of new experimental API reference pages has been published under the \experiments-api\ section, documenting new capabilities such as tool-using agents with human-in-the-loop confirmation strategies, in-memory and Mem0-backed chat message stores, hallucination risk scoring for OpenAI chat generators, and components for summarization, preprocessing, retrieval, and writing chat messages.
_docs-website/reference\_versioned\docs/version-2.27 · high confidence
Promotes Haystack 2.30 documentation and adds experimental API references
The version-2.30 documentation site is now promoted as the stable release, making it the primary reference for users. This update also introduces a new set of experimental API reference pages covering components such as tool-using agents with human-in-the-loop confirmation, in-memory chat message storage, Mem0-based memory stores, LLM-powered summarizers, and hallucination risk scoring for OpenAI chat generators.
_docs-website/reference\_versioned\docs/version-2.30 · high confidence
Promotes Haystack 2.31 experimental APIs to stable reference documentation
The Haystack 2.31 reference documentation now includes stable API references for previously experimental components. This update adds comprehensive documentation for the core Agent component (including Jinja2 user prompt templates and confirmation strategies), the ChatMessageStore ecosystem (InMemory, Mem0, ChatMessageRetriever, and ChatMessageWriter), and new experimental capabilities such as the OpenAIChatGenerator with hallucination risk scoring, the LLMSummarizer, and the MarkdownHeaderLevelInferrer. Also in this area: 9 commits (1 fix) modifying docs-website; 12 commits (3 fixes) modifying docs-website.
_docs-website/reference\_versioned\docs/version-2.31 · medium confidence · unverified
TopPSampler now supports a minimum document count via min\_top\_k
The TopPSampler component in haystack/components/samplers now accepts a min\_top\_k parameter. This allows users to specify a minimum number of documents to return, ensuring that even if top-p sampling selects fewer documents than expected, the result will include at least min\_top\_k documents with the highest scores. The component also validates that min\_top\_k is a non-negative integer or None, raising a ValueError if invalid.
haystack/components/samplers · high confidence
Removals
Removal of legacy FARM-based question answering components
The \farm\_haystack\ module has been completely removed, eliminating the legacy FARM-based question answering stack. This deletion removes the \Finder\ orchestration class, the \FARMReader\ and \TfidfRetriever\ implementations, the Flask-based REST API (\api/inference.py\), the SQL database ORM and configuration, and the document indexing and cleaning utilities. Users relying on this specific FARM integration for document retrieval and answer generation will no longer have access to these components.
_farm\haystack · high confidence
Behavioural changes
1 commit (0 fixes) modifying haystack/hooks/backup\_answer
A change to existing behaviour in haystack/hooks/backup\_answer — 1 commit, 2 files: “feat: add experimental BackupAnswerHook (\#13553)” (refs \#13553).
_haystack/hooks/backup\answer · medium confidence · unverified
32 commits (0 fixes) modifying docs-website/reference\_versioned\_docs/version-3.4
A change to existing behaviour in docs-website/reference\_versioned\_docs/version-3.4 — 32 commits, 139 files: “docs: sync Core Integrations API reference (serperdev) on Docusaurus (\#13330)”; “docs: sync Core Integrations API reference (lara) on Docusaurus (\#13324)”; “docs: sync Core Integrations API reference (anthropic) on Docusaurus (\#13316)”; and 29 more (refs \#13330, \#13324, \#13316, \#13327, \#13367, …). Also in this area: 9 commits (1 fix) modifying docs-website.
(repo-wide) · medium confidence · unverified
Add versioned API reference documentation for Haystack 2.19
A new API reference page has been added for version 2.19, providing a structured overview of the Haystack framework's technical documentation. This entry point organizes the reference material into three distinct sections: the core Haystack API (covering base components, pipelines, document stores, and utilities), the Integrations API (for official packages connecting to external services), and the Experiments API (for features under active development).
_docs-website/reference\_versioned\docs/version-2.19 · high confidence
Code blocks now hide language badges and action buttons based on line count
The documentation website now dynamically adjusts the appearance of code blocks to reduce visual clutter for short snippets. Single-line code blocks will no longer display the language badge, and two-line code blocks will hide the copy and other action buttons, while multi-line blocks retain the full interface.
docs-website/src/theme/CodeBlock · high confidence
Docs now automatically version cross-plugin internal links
A new Remark plugin (\versionedReferenceLinks\) has been added to the documentation build process to automatically inject version prefixes into internal links between the main docs and the reference section. This ensures that when users navigate from a specific documentation version (e.g., v2.19) to the reference, they are directed to the corresponding versioned reference page (e.g., \/reference/v2.19/...\) rather than the latest version. The latest version remains unversioned, while the current/next version uses a \/next/\ prefix, preventing broken links when users browse older documentation versions.
docs-website/src/remark · high confidence
Documentation for Haystack 2.27 pipeline components
The documentation for the Haystack 2.27 release has been updated to include new and refactored guides for pipeline components. This includes new pages for the \Agent\ component (covering tool usage, multi-agent systems, and streaming) and the \Human-in-the-Loop\ system (detailing confirmation strategies, policies, and UIs). Audio processing documentation has been added for \LocalWhisperTranscriber\ and \RemoteWhisperTranscriber\, alongside updated guides for builders like \ChatPromptBuilder\ and \PromptBuilder\, and classifiers such as \DocumentLanguageClassifier\.
_docs-website/versioned\docs/version-2.27/pipeline-components · high confidence
Documentation site assets and configuration updates
The documentation website now includes a .nojekyll file to ensure proper static file serving on GitHub Pages. New SVG icons for the chevron, copy-to-clipboard, and Discord community have been added to the static image directory. Additionally, a new 'Vector Databases' overview diagram has been introduced to help users understand document store architectures.
docs-website/static · high confidence
Haystack 2.26 documentation site goes live with new overview pages
The versioned documentation for Haystack 2.26 is now available, introducing a set of new overview pages that were previously missing or unstable. Users can now find official guides on the breaking change policy, installation via pip/uv/conda, and a comprehensive migration guide from Haystack 1.x. The release also includes a new FAQ page, a telemetry policy page explaining anonymous usage statistics and opt-out methods, and an extended migration guide for users moving from LangChain/LangGraph. Additionally, the 'Get Started' page has been updated with new code examples for building RAG pipelines and agents using OpenAI, Hugging Face, and Anthropic.
_docs-website/versioned\docs/version-2.26/overview · high confidence
Haystack 3.3 documentation release with new agent and component concepts
The versioned documentation for Haystack 3.3 has been updated to reflect the release of new core concepts and structures. This includes new pages explaining the Agents framework (including multi-agent systems and choosing between pipelines and agents), SuperComponents for wrapping pipelines as single components, and custom component creation. The update also introduces standardized templates for component and document store documentation, updates data class documentation to include new types like FileContent and ImageContent, and adds a CLAUDE.md file for agent configuration.
_docs-website/versioned\docs/version-3.3 · high confidence
Haystack package restructure and logging overhaul
The Haystack package has been restructured, introducing a new \haystack/\_\init\\_.py\ that explicitly exports core primitives like \Pipeline\, \Document\, and \Answer\, alongside a new \SuperComponent\ decorator. A significant behavioral change is the introduction of a dedicated logging module (\haystack/logging.py\) that automatically configures logging upon import, enforcing keyword-only arguments for log calls and capping field sizes to prevent log pipeline stalls. The package also now includes a \py.typed\ marker file to signal full type completeness and a new \lazy\_imports.py\ utility for handling optional dependency errors with clearer messages.
haystack · high confidence
In-memory retrievers now support configurable filter policies
The \InMemoryBM25Retriever\ and \InMemoryEmbeddingRetriever\ components now accept a \filter\_policy\ parameter (defaulting to \REPLACE\). This allows users to choose between \REPLACE\, which overrides initialization filters with runtime filters, and \MERGE\, which combines them to narrow the search space. This change ensures that filter application behavior is explicit and consistent, addressing previous issues where the \MERGE\ policy was not applied correctly.
_haystack/components/retrievers/in\memory · high confidence
Introduce Component I/O socket representation and repr
The \haystack.core.component\ module now exposes a \Sockets\ class that provides a structured, human-readable representation of a Component's inputs and outputs. Users can inspect component interfaces via the new \\_\repr\\_\ methods on \Sockets\, \InputSocket\, and \OutputSocket\, which display socket names and types in a clear format (e.g., 'Inputs: - question: Any'). This change improves debugging and introspection capabilities by making component I/O definitions more visible and consistent.
haystack/core/component · high confidence
Introduce lazy loading for haystack.dataclasses
The \haystack.dataclasses\ module now uses a lazy importer to defer the loading of its submodules (such as \answer\, \chat\_message\, \streaming\_chunk\, etc.) until their contents are actually accessed. This change is designed to optimize import times by reducing the initial overhead of loading all dataclass definitions at startup, while still exposing the full public API (including \ChatMessage\, \StreamingChunk\, \Document\, and others) to users.
haystack/dataclasses · high confidence
Introduce lazy loading for the in-memory document store
The in-memory document store module now uses lazy imports to defer loading the \InMemoryDocumentStore\ class until it is actually accessed. This change reduces initial import overhead, improving startup performance for applications that do not utilize the in-memory document store.
_haystack/document\_stores/in\memory · high confidence
Introduce pinned, multi-platform base Docker image for Haystack 2.x
The Docker build system has been restructured to provide a new \haystack:base\ image that users can derive their own images from. This base image is built using \docker buildx bake\, supports multi-platform builds (linux/amd64 and linux/arm64), and pins the Python 3.12-slim base image and the \uv\ installation tool by SHA-256 digest for reproducible, supply-chain-safe builds. The image includes a pre-installed virtual environment with Haystack installed from source, and explicitly upgrades \setuptools\ to address [CVE redacted].
docker · high confidence
New API Reference homepage for Haystack
A new landing page for the API documentation has been added to the reference section. This page serves as the entry point for technical reference material, clearly separating the core framework API (covering base components, pipelines, document stores, and utilities) from the Integrations API (covering official packages that connect Haystack to external services).
docs-website/reference · high confidence
New and updated document store documentation for Haystack 2.26
The documentation for the \docs-website/versioned\_docs/version-2.26/document-stores\ area has been updated to include new guides for ArcadeDB, Astra, Azure AI Search, FAISS, InMemory, MongoDB Atlas, Pgvector, Pinecone, Qdrant, Valkey, and Weaviate document stores. The OpenSearch guide has been updated to reflect version 3.5.0, and the Elasticsearch guide now specifies support for version 8.11.1. These changes provide users with installation instructions, initialization examples, and supported retriever details for these specific integrations.
_docs-website/versioned\docs/version-2.26/document-stores · high confidence
New documentation for Haystack 2.29 concepts, templates, and data classes
The documentation site for version 2.29 has been updated with new concept pages for Agents, Multi-Agent Systems, Components, and SuperComponents, alongside dedicated documentation for the FileContent data class and new MDX templates for component and document store pages.
_docs-website/versioned\docs/version-2.29 · high confidence
New search modal with product-aware content filtering
The documentation site now features a new search modal (replacing the previous inline search bar) that supports filtering by 'All', 'Documentation', and 'API Reference'. The search experience is product-aware: it detects whether the user is viewing the Professional or Enterprise documentation (via URL path) and automatically hides Table of Contents items and search results that do not belong to the current product context. The modal also includes UI improvements such as branded styling, single-word sidebar link truncation, and a 'Powered by Haystack' attribution.
docs-website/src/theme · high confidence
Promote Haystack 2.26 documentation to stable reference
The documentation for Haystack version 2.26 has been promoted from unstable to stable, making the full API reference available for this release. This update includes the complete experimental API documentation—covering Agents, ChatMessage Stores, Generators, Memory Stores, Preprocessors, Retrievers, Summarizers, and Writers—as well as the stable API reference for Agents and Audio components.
_docs-website/versioned\docs/version-2.22/pipeline-components · high confidence
Promoted Haystack 2.31 documentation with new agent and component guides
The documentation for Haystack version 2.31 has been promoted to stable, introducing comprehensive guides for building AI agents and multi-agent systems, including coordinator/specialist architectures. New concept pages explain core building blocks like Components, SuperComponents, and Data Classes (such as ChatMessage and FileContent), while new templates standardize component documentation. The release also updates installation instructions to use the \weave-haystack\ package and removes documentation for components deprecated in version 3.
_docs-website/versioned\docs/version-2.31 · high confidence
Promoted Haystack 3.3 reference documentation to stable
The Haystack 3.3 API reference documentation is now stable and available in the versioned docs. This update includes comprehensive reference pages for core components such as Agents, Builders, Caching, Converters, Data Classes, Document Stores, Document Writers, Embedders, Evaluation, and Evaluators, providing users with finalized API signatures, usage examples, and parameter details for these features.
_docs-website/reference\_versioned\docs/version-3.1 · high confidence
Standardize e2e test environment with telemetry blocking and HF credential cleanup
The e2e test suite now includes a centralized conftest that automatically disables telemetry network calls for all tests and removes Hugging Face environment variables (HF\_API\_TOKEN, HF\_TOKEN) before test execution. This prevents accidental API calls and failures caused by empty tokens, particularly in fork-based pull requests, while ensuring consistent test isolation.
e2e · high confidence
Telemetry module restructured with PostHog integration and updated event schema
The telemetry subsystem has been reorganized into a new package structure (haystack/telemetry) with dedicated modules for environment detection and core telemetry logic. System specification collection now explicitly excludes transformers and torch libraries, setting their values to False by default to reduce noise. Telemetry events are now sent to PostHog using a specific API key, and the pipeline run event version has been bumped to 'Pipeline run (3.x)'. The implementation also fixes a ResourceWarning regarding unclosed files and ensures the telemetry decorator preserves original function metadata.
haystack/telemetry · high confidence
Updated documentation site styling with Inter font and new color palette
The documentation website now uses the Inter font family for all text and headings, replacing the previous default. A new blue-based color palette has been applied across light and dark modes, affecting primary/secondary colors, admonition backgrounds and borders, and neutral UI components like buttons and breadcrumbs. The active state for navbar items (Docs and API Reference) is now highlighted in the primary blue color with increased font weight, and breadcrumb links follow a similar active-state styling convention.
docs-website/src/css · high confidence
Updated introduction page with rebranded messaging and enterprise promotion
The version 2.21 documentation now includes a new introduction page that rebrands Haystack as an open-source AI framework for building production-ready AI Agents, RAG applications, and multimodal search systems. The content has been updated to reflect the rebranding, correct typos, and promote the Haystack Enterprise Starter and Platform for teams seeking enterprise-grade support and scalable tooling.
_docs-website/versioned\docs/version-2.21 · high confidence
Fixes
5 commits (4 fixes) fixing haystack/hooks/structured\_output
A fix in haystack/hooks/structured\_output — 5 commits (4 fixs), 2 files: “fix: mark user messages added by hooks as system reminders (\#13580)”; “feat: add StructuredOutputHook to parse an Agent's final reply into structured output (\#13240)”; “fix: don't let system reminders reset StructuredOutputHook retries (\#13501)”; and 2 more (refs \#13580, \#13240, \#13501, \#13400, \#13617).
_haystack/hooks/structured\output · medium confidence · unverified
Test coverage
Add comprehensive test suite for ChatPromptBuilder; Added end-to-end tests for RAG evaluation and PDF content extraction pipelines; Added sample components for testing; Added test helper for callable deserialization; New testing utilities and fixtures for document stores and telemetry.
Dependencies
Migrate build system to pyproject.toml and update docs-website dependencies
The project has replaced the legacy requirements.txt with a comprehensive pyproject.toml, establishing a new build system using Hatchling and defining core dependencies such as openai\>=2.6.0, httpx, and anyio\>=4.14.2 (pinned to address [CVE redacted]). The docs-website now uses a dedicated package.json featuring Docusaurus 3.10.2, React 19, and Vercel 62.1.0, while removing the old requirements.txt which previously listed deprecated packages like FARM and Flask.
(dependencies) · high confidence
Housekeeping
Add Apache 2.0 license header to document stores module; Documentation for Haystack 2.21 pipeline components.
Written by cca.canine.dev from the codebase's own history, inside the signed delivery this page is composed from.
How this codebase got here
Score
- CAI 66.7 → 71.3 (+4.6)
- Rubric changed (rubric-2026.11.2 → rubric-2026.11.4) — scores are not directly comparable.
Lenses
- Code Health 85.9 → 82.3 (-3.6)
- Architecture 86.0 → 96.3 (+10.3)
- Maturity 75.9 → 79.8 (+3.9)
- Readiness 80.9 → 80.7 (-0.2)
- Security 79.2 → 79.2 (0.0)
- Accessibility 51.8 → 59.7 (+7.9)
- Performance 100.0 → 100.0 (0.0)
Resolved (17)
- Duplicated block (11 lines × 2) (haystack/components/retrievers/auto_merging_retriever.py)
- Duplicated block (6 lines × 2) (haystack/components/agents/utils.py)
- Duplicated block (7 lines × 2) (haystack/components/routers/conditional_router.py)
- Hotspot: haystack/components/agents/agent.py (haystack/components/agents/agent.py)
- Hotspot: haystack/components/preprocessors/recursive_splitter.py (haystack/components/preprocessors/recursive_splitter.py)
- Hotspot: haystack/components/routers/file_type_router.py (haystack/components/routers/file_type_router.py)
- Hotspot: haystack/hooks/human_in_the_loop/strategies.py (haystack/hooks/human_in_the_loop/strategies.py)
- Off-boarding risk: anonymized user #1
- Projects may be oversized for their cohesion
- PythonCodeSplitter._emit_class_units (cyclomatic 19) (haystack/components/preprocessors/python_code_splitter.py)
- Repeated repair: haystack/components/converters/output_adapter.py (haystack/components/converters/output_adapter.py)
- Repeated repair: haystack/components/evaluators/document_map.py (haystack/components/evaluators/document_map.py)
- Repeated repair: haystack/utils/callable_serialization.py (haystack/utils/callable_serialization.py)
- Scanner failed to run — not a clean result
- Scanner failed to run — not a clean result
- generate_platform_components_table.scan_for_components (cognitive 16) (scripts/generate_platform_components_table.py)
- openai_responses._convert_response_to_chat_message (cyclomatic 17) (haystack/components/generators/chat/openai_responses.py)
New (194)
- Duplicated block (10 lines × 2) (haystack/components/agents/tool_calling.py)
- Duplicated block (13 lines × 2) (haystack/components/generators/chat/openai.py)
- Duplicated block (13 lines × 2) (haystack/components/generators/chat/openai.py)
- Duplicated block (16 lines × 2) (haystack/components/retrievers/auto_merging_retriever.py)
- Duplicated block (6 lines × 2) (haystack/hooks/backup_answer/hooks.py)
- Duplicated block (6 lines × 2) (haystack/hooks/backup_answer/hooks.py)
- Duplicated block (86–101 lines × 2) (haystack/components/generators/chat/openai.py)
- Edited copies of one member (3 members, 10+ corresponding lines) (haystack/dataclasses/byte_stream.py)
- Edited copies of one member (3 members, 15+ corresponding lines) (haystack/components/embedders/openai_text_embedder.py)
- Edited copies of one member (3 members, 25+ corresponding lines) (haystack/testing/document_store.py)
- Edited copies of one member (3 members, 49+ corresponding lines) (haystack/document_stores/in_memory/document_store.py)
- Edited copies of one member (3 members, 6+ corresponding lines) (haystack/components/generators/chat/openai.py)
- Edited copies of one member (3 members, 64+ corresponding lines) (haystack/components/generators/chat/openai.py)
- Edited copies of one member (4 members, 19+ corresponding lines) (haystack/components/routers/openai_decisions_image_router.py)
- Edited copies of one member (4 members, 19+ corresponding lines) (haystack/testing/document_store.py)
- Edited copies of one member (4 members, 29+ corresponding lines) (haystack/components/retrievers/in_memory/bm25_retriever.py)
- Edited copies of one member (4 members, 9+ corresponding lines) (haystack/components/converters/pypdf.py)
- Edited copies of one member (5 members, 6+ corresponding lines) (haystack/components/extractors/llm_metadata_extractor.py)
- Edited copies of one member (5 members, 6+ corresponding lines) (haystack/testing/document_store.py)
- Edited copies of one member (6 members, 6+ corresponding lines) (haystack/components/extractors/llm_metadata_extractor.py)
- …and 174 more
Changes since last survey
- 156 commits — 94 feature/other, 62 fixes
By area
- docs-website/reference — 33 commits
- haystack/components — 33 commits
- docs-website/docs — 29 commits
- docs-website/reference_versioned_docs — 22 commits
- .github/workflows — 7 commits
- haystack/document_stores — 6 commits
- test/components — 5 commits
- docs-website/versioned_docs — 3 commits
- haystack/tools — 3 commits
- haystack/utils — 3 commits
- e2e/pipelines — 2 commits
- haystack/core — 2 commits
- haystack/dataclasses — 2 commits
- test/core — 2 commits
- (root) — 1 commit
- docs-website/package.json — 1 commit
- haystack/hooks — 1 commit
- test/tools — 1 commit
Notable commits
- fix: fix!: pass variadic inputs in connection order in Pipeline.run() and run_async() (#13415)
- fix: fix(compaction): keep another strategy's summary in sliding window (#13375)
- fix: fix(pptx): keep soft line breaks with markdown and plain link formats (#13222)
- fix: fix: DocumentToImageContent skips undecodable images instead of failing the whole batch (#13452)
- fix: fix: accept messages from other providers in OpenAIResponsesChatGenerator (#13526)
- fix: fix: accept signed offsets in the Jinja2 {% now %} tag (#13653)
- fix: fix: add delete_by_filter_async to InMemoryDocumentStore (#13356)
- fix: fix: allow reserved names in tool output mappings (#13159)
- fix: fix: apply each TextCleaner remove_regexps pattern independently (#13293)
- fix: fix: avoid mutating MetadataRouter deserialization input (#13421)
- fix: fix: avoid repeated text after contained sentence chunks (#13340)
- fix: fix: complete partial dates to start of period in range filters (#13342)
- fix: fix: convert common Python types in Agent tool results to JSON (#13728)
- fix: fix: deduplicate AutoMergingRetriever documents (#13511)
- fix: fix: don't count whitespace-only pieces as words in RecursiveDocumentSplitter (#13354)
- fix: fix: don't let system reminders reset StructuredOutputHook retries (#13501)
- fix: fix: honor runtime streaming callback precedence (#13405)
- fix: fix: index snake_case-split tool name in SearchableToolset BM25 (#13633)
- fix: fix: keep code that shares a line with a docstring in PythonCodeSplitter (#13517)
- fix: fix: keep in-memory BM25 statistics coherent during concurrent mutations (#13674)
- …and 136 more
Architecture
- Unchanged — 0 containers · 1 contexts · 0 edges
Written by cca.canine.dev from the codebase's own history, inside the signed delivery this page is composed from.
Survey your own repository
deepset-ai/haystack was measured the same way every project in this corpus was: the same rubric, at a pinned commit, with the result published in full. Point a surveyor at a repository you know and see whether you agree with it.
About this page
- The score is its most recent published measurement, taken on 16 November 2026 at a pinned commit. It is not a live figure and does not change until the project is measured again.
- Measured at commit 3bcaec8feb5af71ffb925282dace45954661607b — the exact code this score is about.
- Scored under rubric-2026.11.4 — the same rubric and the same method as every other entry in this index.
- Measured by Canine Code Analyzer (cca.canine.dev) using cca measure-2026.11.83+preprod-a489942cf94b.