{"full_name":"deepset-ai/haystack","owner":"deepset-ai","name":"haystack","shard":"1","url":"https://github.com/deepset-ai/haystack","description":"Open-source AI orchestration framework for building context-engineered, production-ready LLM applications. Design modular pipelines and agent workflows with explicit control over retrieval, routing, memory, and generation. Built for scalable agents, RAG, multimodal applications, semantic search, and conversational systems.","homepage":"https://haystack.deepset.ai","topics":["semantic-search","information-retrieval","ai","python","large-language-models","generative-ai","llm","rag","retrieval-augmented-generation","agents","orchestration","agent-framework","agentic-ai","agentic-rag","ai-agents"],"tag_pages":["agent-framework","agentic-ai","agentic-rag","agents","ai","ai-agents","generative-ai","information-retrieval","large-language-models","llm","orchestration","python","rag","retrieval-augmented-generation","semantic-search"],"language":"Python","license":"Apache-2.0","avatar":"https://avatars.githubusercontent.com/u/51827949?v=4","created_at":"2019-11-14T09:05:28Z","pushed_at":"2026-11-14T02:11:02Z","metrics":{"stars":27010,"forks":3469,"watchers":164,"issues":4683,"prs":7907,"releases":256,"issues_open":98,"issues_closed":4585,"prs_open":67,"prs_closed":1139,"prs_merged":6701,"discussions":868},"growth":{"day":{"stars":4,"forks":4,"base_day":"2026-11-15"},"week":{"stars":89,"forks":82,"base_day":"2026-11-04"}},"history":{"dates":["2026-10-20","2026-10-21","2026-10-22","2026-10-23","2026-10-24","2026-10-25","2026-10-26","2026-10-27","2026-10-28","2026-10-29","2026-10-30","2026-10-31","2026-11-01","2026-11-02","2026-11-03","2026-11-04","2026-11-10","2026-11-11","2026-11-12","2026-11-13","2026-11-14","2026-11-15","2026-11-16"],"stars":[26744,26746,26757,26777,26792,26804,26812,26818,26826,26847,26859,26871,26876,26886,26913,26921,26975,26990,26997,26997,27003,27006,27010],"forks":[3286,3297,3300,3309,3306,3316,3321,3332,3333,3345,3349,3353,3359,3368,3384,3387,3424,3445,3452,3453,3457,3465,3469],"watchers":[166,166,167,167,167,167,167,167,164,164,164,164,164,164,164,164,null,165,165,165,165,164,164],"issues_open":[null,null,null,null,null,null,null,null,null,null,null,null,null,null,null,null,null,89,89,89,92,97,98],"prs_open":[null,null,null,null,null,null,null,null,null,null,null,null,null,null,null,null,null,58,55,59,53,62,67]},"boards":{},"same_owner":[],"issues":[{"number":11039,"title":"RFC: Signed receipts for Haystack pipeline component calls","comments":20,"created_at":"2026-04-05T14:52:48Z"},{"number":10810,"title":"EU AI Act compliance scan results — Haystack scored #1, wanted to validate findings","comments":17,"created_at":"2026-03-12T20:50:00Z"},{"number":11836,"title":"feat: Pipeline run recording & deterministic replay - turn every run into a shareable, diffable, testable artifact","comments":10,"created_at":"2026-06-30T14:05:07Z"},{"number":10973,"title":"Feature: Post-RAG FactChecker Pipeline Component & Header-Aware Document Splitter","comments":9,"created_at":"2026-03-29T19:06:34Z"},{"number":10821,"title":"Feature: GuardrailProvider interface for automated tool-call policy enforcement","comments":9,"created_at":"2026-03-14T18:08:03Z"},{"number":11266,"title":"Proposal: Transaction Protocol for idempotent, auditable agent pipelines","comments":7,"created_at":"2026-05-06T12:59:17Z"},{"number":6970,"title":"Streamlit Stream Callback in Haystack 2.x","comments":7,"created_at":"2024-02-11T18:41:06Z"},{"number":12060,"title":"Allow Human-in-the-Loop to review fully materialized tool calls","comments":6,"created_at":"2026-07-18T11:54:44Z"}],"discussions":[{"number":13705,"title":"mcp-slim: Zero-dependency MCP stdio proxy for Haystack MCPToolset (-50% to -98% schema tokens & SecretGuard)","comments":2,"category":"Show and tell","created_at":"2026-11-12T02:35:33Z"},{"number":13626,"title":"OpenAIChatGenerator against a non-OpenAI base URL: api_base_url vs OPENAI_BASE_URL, stacked retries, and keeping keys out of pipeline YAML","comments":0,"category":"Show and tell","created_at":"2026-11-09T03:43:28Z"},{"number":13609,"title":"Testing the scope of remembered tool approval: ARQUIA-EVAL-001 open fixture","comments":0,"category":"Show and tell","created_at":"2026-11-07T01:29:36Z"},{"number":13584,"title":"ArtiGram: a free gallery for AI-made interactive artifacts, with a remote MCP server","comments":0,"category":"Show and tell","created_at":"2026-11-06T08:29:11Z"},{"number":13407,"title":"Testing error propagation from MCPToolset through a two-component Haystack pipeline","comments":0,"category":"Show and tell","created_at":"2026-10-23T10:20:53Z"},{"number":13205,"title":"A Haystack component that prepares a payment request instead of paying: agent proposes, human signs","comments":0,"category":"Show and tell","created_at":"2026-10-10T18:12:58Z"},{"number":13048,"title":"Per-run dollar budgets for Haystack pipelines and agents (no custom component)","comments":0,"category":"Show and tell","created_at":"2026-10-01T01:55:41Z"},{"number":12969,"title":"Your retrieval budget may be based on a context window you are not actually being served","comments":1,"category":"General","created_at":"2026-09-26T02:18:10Z"}],"latest_release":{"name":"v3.5.0","tag":"v3.5.0","published_at":"2026-11-10T12:34:55Z","body":"## ⭐️ Highlights\n\n### 🧭 Classify and Route with the OpenAI Decisions API\nThree new components use the OpenAI Decisions API to choose how to handle your data without instructions. [`OpenAIDecisionsTextRouter`](https://docs.haystack.deepset.ai/docs/openaidecisionstextrouter) sends text to a named output per label, [`OpenAIDecisionsImageRouter`](https://docs.haystack.deepset.ai/docs/openaidecisionsimagerouter) does the same for images and can send low-confidence decisions to a fallback branch, and [`OpenAIDecisionsDocumentClassifier`](https://docs.haystack.deepset.ai/docs/openaidecisionsdocumentclassifier) answers choice, score, and yes/no questions about documents and stores the answers in their metadata.\n\n```python\nfrom haystack.components.routers import OpenAIDecisionsTextRouter\n\nrouter = OpenAIDecisionsTextRouter(\n    labels={\n        \"billing\": \"Payments, invoices, refunds and charges\",\n        \"technical\": \"Bugs, crashes and errors in the product\",\n    },\n    instructions=\"Which team should handle this support request?\",\n    min_confidence=0.6,\n)\n\nprint(router.run(text=\"I was charged twice for my subscription.\"))\n# >> {'billing': 'I was charged twice for my subscription.'}\n```\n\n### 🧱 Structured Output for Agents\nThe new [`StructuredOutputHook`](https://docs.haystack.deepset.ai/docs/structured-output) validates the Agent's final reply against a Pydantic model or JSON schema and returns it as a dict in `result[\"structured_output\"]`. If the reply doesn't validate, the hook sends the error back to the model and lets it try again, up to `max_retries` times.\n\n```python\nfrom pydantic import BaseModel\n\nfrom haystack.components.agents import Agent\nfrom haystack.components.generators.chat import OpenAIResponsesChatGenerator\nfrom haystack.dataclasses import ChatMessage\nfrom haystack.hooks.structured_output import StructuredOutputHook\n\nclass CityInfo(BaseModel):\n    city: str\n    country: str\n\nagent = Agent(\n    chat_generator=OpenAIResponsesChatGenerator(generation_kwargs={\"text_format\": CityInfo}),\n    state_schema={\"structured_output\": {\"type\": dict}},\n    hooks={\"on_exit\": [StructuredOutputHook(schema=CityInfo)]},\n)\n\nresult = agent.run(messages=[ChatMessage.from_user(\"Which city is the Eiffel Tower in?\")])\nprint(result[\"structured_output\"])\n# >> {'city': 'Paris', 'country': 'France'}\n```\n\nAlso new for Agents: the experimental [`StepBudgetReminderHook`](https://docs.haystack.deepset.ai/docs/step-budget-reminder) tells the model how many steps it has left as it approaches `max_agent_steps`, so it wraps up with an answer instead of being cut off.\n\n### 🎯 Multivector Embeddings and MaxSimRanker\n`Document` has a new `multivector_embedding` field for late-interaction models such as ColBERT, and the new `MaxSimRanker` reranks candidates by comparing each query vector to the document vectors and summing the best matches. `DocumentJoiner(join_mode=\"merge_embeddings\")` can also combine dense, sparse, and multivector embeddings computed in parallel for the same documents.\n\nThis is the first step toward multivector retrieval in Haystack ([#13257](https://github.com/deepset-ai/haystack/issues/13257)). Embedders that produce multivector embeddings are in progress in Haystack Core Integrations: [Sentence Transformers](https://github.com/deepset-ai/haystack-core-integrations/pull/4208) and [vLLM](https://github.com/deepset-ai/haystack-core-integrations/pull/4014).\n\n```python\nfrom haystack import Document\nfrom haystack.components.rankers import MaxSimRanker\n\n# Representative multivector embedding for \"How do solar panels work?\"\nquery_multivector_embedding = [[1.0, 0.0], [0.0, 1.0]]\n\ndocuments = [\n    Document(content=\"Wind turbines generate electricity.\", multivector_embedding=[[1.0, 0.0], [0.0, 0.2]]),\n    Document(content=\"Solar panels turn sunlight into power.\", multivector_embedding=[[0.95, 0.05], [0.15, 0.9]]),\n    Document(content=\"Solar cells absorb light.\", multivector_embedding=[[0.9, 0.1], [0.1, 0.9]]),\n]\n\nranker = MaxSimRanker(top_k=2)\nre","is_prerelease":false},"readme":true,"readme_zh":false,"news":[],"stale":false,"snapshot_at":"2026-11-16T09:23:14Z","history_start":"2026-09-06"}