deepset: Open-Source IDP and RAG Orchestration Framework

deepset is a Berlin-based AI company founded in 2018 by Milos Rusic (CEO), Malte Pietsch, and Timo Möller. The company builds Haystack, an open-source Python framework for constructing retrieval-augmented generation (RAG) pipelines, AI agents, and intelligent document processing (IDP) workflows. Haystack carries an Apache 2.0 license and has approximately 25,000 GitHub stars. The company employs 51-200 people and is privately held.

Identity and ownership

deepset is an independent private company. No acquisition, parent company, or external ownership change has been disclosed as of June 2026. Milos Rusic has served as CEO since founding. The company holds SOC 2 Type II, ISO 27001, GDPR, and HIPAA certifications.

Gartner named deepset a Cool Vendor in AI Engineering in 2024, per the report authored by Arun Chandrasekaran, Manjunath Bhat, Arup Roy, and George Brocklehurst. "Cool Vendor" is an early-stage recognition category, not a Magic Quadrant Leader or Challenger position.

Products

deepset's three-tier structure mirrors the open-source monetization model used by other infrastructure companies: a free tier drives developer adoption, commercial tiers capture enterprises at scale.

Pricing for Enterprise Starter is based on organization size with no per-seat or usage-based model disclosed. The rebrand from "deepset AI Platform" to "Haystack Enterprise Platform" in early 2026 unified the open-source and commercial identities under a single name, resolving a long-standing positioning ambiguity.

How Haystack processes documents

Haystack uses a modular directed acyclic graph (DAG) pipeline architecture. Each processing step, including document ingestion, chunking, embedding, retrieval, reranking, and generation, is a serializable, swappable component. Pipelines are defined in Python and exported to YAML for reproducible deployment across Docker, Kubernetes, Ray, and serverless environments.

For RAG workflows, Haystack connects to 80+ data sources and model providers including Hugging Face, Elasticsearch, OpenSearch, OpenAI, Cohere, and Anthropic. Built-in evaluation tooling measures retrieval quality and generation accuracy before pipelines reach production. As futureagi.com noted in April 2026, the typed pipeline graph with explicit connections is more opinionated than LangChain's chain abstraction, which creates a steeper initial learning curve but more predictable production behavior.

The framework emits OpenTelemetry-compatible spans through a native tracing module, plus OpenInference and traceAI instrumentation packages, giving operations teams visibility into pipeline execution.

The 2024 architectural rewrite (Haystack 2.x) introduced typed component sockets, explicit directed-graph pipeline semantics with branching and cycles for agent loops, an async-friendly runtime, and YAML-serialized pipelines for declarative deployment. The Agent component fits natively inside a Pipeline alongside retrievers and generators.

Recent releases extend the architecture in concrete directions. Haystack v2.29.0 introduced MultiRetriever, which runs multiple text retrievers in parallel and merges results using reciprocal rank fusion by default. A companion TextEmbeddingRetriever enables hybrid BM25-plus-embedding search pipelines, directly relevant to IDP use cases where keyword precision and semantic recall both matter, such as extracting specific clauses from legal documents. Haystack v2.30.0 added PythonCodeSplitter, a syntax-aware component that parses Python source files using Python's ast module and keeps whole functions together within configurable chunk sizes. Haystack v2.28.0 migrated HTTP handling from requests to httpx, standardizing synchronous and asynchronous request patterns, and added State parameter injection in tool signatures so tools can access live agent state without explicit wiring. A v2.30.2 patch fixed an Agent exit condition bug where the agent would exit prematurely if the LLM produced invalid tool calls.

IDP use cases and customer outcomes

deepset's IDP documentation describes three production deployments in financial services. The human-in-the-loop validation pattern deepset describes, where analysts verify extracted information before report generation, positions Haystack as supervised automation rather than fully autonomous processing. This is a credible positioning for regulated industries where full straight-through processing (STP) carries compliance risk.

deepset identifies finance, insurance, healthcare, and legal as primary target verticals due to document-heavy workflows requiring in-depth analysis. No independent third-party verification of these outcomes has been published as of June 2026.

Sovereign AI and government positioning

Three developments mark deepset's clearest strategic direction. Germany's Federal Ministry for Digital Affairs selected deepset to run 18 AI agent pilots across 17 municipalities, covering document analysis and administrative process automation. Haystack was chosen as a reference architecture for Deutschland-Stack (D-Stack), Germany's national sovereign technology platform, with secunet Security Networks AG announcing a reference architecture for processing classified information powered by Haystack. deepset also partnered with Celonis to combine Haystack with Celonis Process Intelligence for a sovereign AI platform targeting regulated sectors including military, cybersecurity, policing, and national infrastructure.

The NVIDIA NIM integration and the Meta Llama Stack partnership both target the same buyer: regulated industries that cannot route documents through public cloud inference endpoints. No major US-headquartered IDP platform has achieved equivalent government validation in Germany, which gives deepset a reference position that competitors cannot easily replicate in European public sector procurement.

Competitive position

deepset competes on three fronts. Against general-purpose LLM frameworks like LangChain and LlamaIndex, deepset emphasizes production-readiness and governance over rapid prototyping, as documented in the DigitalOcean RAG framework comparison. Against monolithic IDP platforms, Haystack offers modular, open-source alternatives that give engineering teams full control over the processing stack. Against in-house engineering efforts, Haystack provides pre-built components, evaluation tooling, and maintainer support that reduce time-to-production.

For teams evaluating open-source document processing frameworks, competing approaches include Docling from IBM Research (MIT license, strong layout preservation) and Unstructured (ETL-focused, 25+ file types). A direct comparison is available in the Unstructured competitive analysis. For broader context on self-hosted options, see the self-hosted document processing guide and the document processing for RAG guide.

Technical specifications

Resources

• Website
• GitHub Repository
• Haystack Documentation
• MCP Server Package
• SingleStore Integration
• Integrations Directory