deepset, makers of Haystack

⚡ We're proud to partner with @mozilla on #Thunderbolt to deliver a complete sovereign AI stack for enterprises and public sector organizations that need full control over how AI is built, deployed, and used. Together, Thunderbolt and @Haystack_AI give organizations a self-hostable AI client paired with the infrastructure to run agents, RAG pipelines, and AI workflows on your organization's actual data. Deployed within your infrastructure, free from vendor lock-in. Learn more about our partnership here: thunderbolt.io/announcing-thu…

AI sovereignty has become a mainstream enterprise priority. But the conversation often centers on a single question: Where does the model run? That's only part of the story. When you deploy AI agents in production, the real value accumulates around the model: the skills that encode how your team works, along with the tools, memory, workflows, and context unique to your organization. Models can be replaced. This agent orchestration layer cannot. The more important question becomes: Who controls that layer? If your skills, logic, and context only work inside a single vendor's runtime, lock-in hasn't disappeared. It has simply moved from the model to the orchestration layer around it. Today we're excited to introduce Haystack 3.0, designed to keep that layer open and under your control: → Portable skills you can read, version, and own → A programmable agent loop you can inspect, audit, and adapt → Built-in observability into how your agents think and operate Every AI agent runs inside an orchestration layer. True AI sovereignty means owning the agent orchestration layer, not just choosing where the model runs. With 3.0, that control stays in your hands: open source, model-agnostic, infrastructure-independent, and designed to keep the most valuable part of your AI system yours. @malte_pietsch dives deeper into this idea in his latest post: deepset.ai/blog/haystack-…

What a search week in Berlin! 🇩🇪 On Tuesday, we hosted "Retrieve. Remember. Reason" at the @deepset_ai offices, organized with @OpenSearchProj and @cognee_, featuring Women of Search (@atitaarora 🙏), where I got to demo what's possible with a knowledge base, memory, and @Haystack_AI's orchestration capabilities. More than half the room was women, which is rare and wonderful at tech meetups! 🫶 Then yesterday, I joined the Vector Space Meetup panel at @aicampusberlin, alongside @qdrant_engine, @cognee_, @llama_index, and @n8n_io, to talk about retrieval strategies, evaluation, orchestration, and what actually goes into production. Great panel and amazing networking session! Kudos to @krotenWanderung and the team It was a great week where I learned so much. Already looking forward to the next thing!

Haystack now publishes a public Model Context Protocol server. Point Claude Code, Cursor, or any MCP-compatible coding agent at it and get real-time access to the latest Haystack docs - no API keys or sign-ups needed. 🖇️ docs.haystack.deepset.ai/docs/docs-mcp-…

We over-trust LLMs, but not worry enough about the agentic loop around them when building agents At my @pyconit talk yesterday, I shared why agents fail and how a solid agent harness can fix them, code & slides below 👇

Can't wait to be back in Bologna for @pyconit this week! I'm talking about why agents fail and how to fix them, with a live demo of our itinerary agent built using context-engineering patterns and @Haystack_AI! See you all there 🙌

WHAT AN EVENT!! 🎉 Last week we hosted an unconference as @deepset_ai with @NVIDIA at @balderton office in London. 🎙️ We kicked off with two short talks. @LukawskiKacper demoed a multi-agentic post-event report creator built with @Haystack_AI and Nemotron 3 Super, @KarinSevegnani walked us through Nemotron 3's performance and benchmarks. 💬 Then we split into three discussion groups and I led a session on memory and long running agents. One of the most interesting conversations I've had in a while. We started with memory and second brain concepts. People shared how they're building personal agents by transcribing YouTube videos and creating knowledge bases following @karpathy's wiki idea. We talked RAG vs filesystems, MCPs vs skills, AI sovereignty and security. At the end of the day, you don't want to lock your whole identity to an API, privacy and freedom of choice matter. Thank you to every attendee for your energy, openness, and contributions. I'm already looking forward to coming back to London 💙 See you at Unconference #4 👀

⭐ @Haystack_AI just crossed 25,000 stars on GitHub! This number means a lot to us, but what it really represents is you. Every contributor who opened a pull request. Every community member who answered a question on Discord. Every developer who filed an issue, wrote a notebook, gave a talk, or simply built something incredible with Haystack and shared it with the world. It's a true community effort 💙 When we first started Haystack, we believed that the journey to building great AI applications should be open, composable, and community-driven. 25,000 stars later, that belief has never felt more validated. Thank you to every single one of you: contributors, users, advocates, and builders. Let’s keep going! 🚀

London people 👋 We're hosting an unconference with @nvidia on April 29th! 💂 This is our third unconference at @deepset_ai and if you haven't been to one before, it's very different from a typical meetup. Instead of talks, there are round table discussions with practitioners working through real context engineering challenges. Expect networking and genuine technical conversations. + we'll be talking about nemotron models and @LukawskiKacper is going to show a live demo. 📅 April 29, 6 PM 📍 @balderton offices Looking forward to the discussions!

Looking forward to @aiDotEngineer next week! 🇪🇺 AI Engineer Paris was one of the best conferences I attended last year, and I'm so excited to be speaking at AI Engineer Europe this year alongside so many brilliant minds. I'll be talking about something that doesn't get enough attention in engineering conversations: sovereignty constraints for technical folks. I'll share real engineering trade-offs and what building under these constraints actually looks like in practice, drawing from my work at @deepset_ai. Hope to see you there, come say hi if you're around! 👋

We hosted the second Unconference in Munich 🇩🇪 with @cognee_, @qdrant_engine , and @deepset_ai, and it was 🔥 The discussions focused on Context Engineering with Open-Source: - How do we get the right context in? - Context in Agentic Systems - Context Engineering in Production …but honestly, we covered everything from automatic ontology generation to coding agents. 💡 One thing I learned: Coding agents often struggle with multiple iterations, partly because MoE models behave differently as context grows. This is why I love hosting and attending these events: you learn from people who are actually building these AI systems and making them work. Huge thanks to everyone who joined 🙌 And kudos to the amazing organizer team: @HandexxK , @krotenWanderung , Paula Plattner 💜 👉 We'll keep on organizing them in different cities. Where should we go next?

🚀 @GoogleDeepMind just launched Gemini Embedding 2 today and Haystack supports it from Day 0. Embed text, images, video, audio, and PDFs into one vector space and build multimodal search or cross-modal retrieval systems. Learn how in our latest blog post 👇 haystack.deepset.ai/blog/multimoda…

🚀 Haystack 2.25 is here! This release focuses on improving the user experience of the Agent component and making tool selection more scalable and efficient. 🛠️ Dynamic Tool Discovery with SearchableToolset With SearchableToolset, agents dynamically discover relevant tools using BM25 keyword search instead of exposing hundreds of tools to the LLM upfront. This approach: – Reduces LLM context usage (and token costs) – Improves tool selection accuracy – Works especially well when connecting to MCP servers Agents load only the tools they actually need, making large tool ecosystems practical without overwhelming the model. 📝 Reusable Jinja2 Prompt Templates for Agents Agents now natively support templated` user_prompt` and `required_variables`, simplifying how you structure and invoke agents. With this update, you can define prompts once and pass variables at runtime. This means: – Less boilerplate – Cleaner agent invocation – More reusable and maintainable agent workflows 💙 Big thanks to our contributors to this release! 👇 Full release notes in the next post Explore the release and let us know what you build with it 💬

Congratulations to @tricalt and the team on raising the $7.5M seed round! Context engineering is one of the most important disciplines in AI, and we’re excited to celebrate this milestone from @cognee_ , pushing the boundaries of context-aware AI systems with memory 🚀

👇 @Haystack_AI continues to enable teams to build more composable, production-ready AI systems, and the 2.24 release focuses on making pipelines simpler and cleaner.

🚀 Haystack 2.23 is out! This release focuses on making agents safer, more visual, and components easier to customize with fewer custom hooks and less boilerplate. Here are the highlights: 🔄 Human-in-the-Loop for Agents Agents can now pause for human confirmation before executing tools. You can define confirmation behavior per tool (always ask, ask once, or never ask), ideal for workflows with sensitive operations. 🖼️ Image Support for Tool Results Tools can now return images alongside text. Agents can retrieve and reason over images, unlocking multimodal workflows like visual search, image inspection, and richer tool outputs. 🧩 Simpler Serialization for Custom Components Custom components now serialize and deserialize automatically in most cases, even when they include complex objects as parameters. This means less boilerplate, easier pipeline snapshots for recovery, and simpler custom component API. 💙 Big thanks to our contributors for the release 👉 Full release notes: haystack.deepset.ai/release-notes/…

🚀 Haystack v2.22 is out! This release brings smarter doc chunking, improved component behavior, and easier tool-to-LLM integration with fewer manual steps. Highlights ✂️ New `EmbeddingBasedDocumentSplitter` for smarter document chunking 🛠️ `outputs_to_string` lets tools return multiple strings with ease 🔥 Components now auto-trigger `warm_up()` on first run 🐍 Python 3.10+ is now required Big thanks to our contributors 💙 Full release notes👇 🔗 haystack.deepset.ai/release-notes/…

Today we’re uniting our product names and look and feel under the @Haystack_AI ecosystem — including renaming the deepset AI Platform to Haystack Enterprise Platform. The platform continues to work exactly as it does today. This update reflects how Haystack brings together open source, enterprise support, and platform capabilities. Details 👇 deepset.ai/blog/introduci…

🦾 @deepset_ai macht produktionsreife KI mit #AWS möglich. 🛠 Die Plattform des #Startups ermöglicht schnellere, skalierbare und sicherere #genAI-Implementierungen und verändert damit die Art und Weise, wie Unternehmen Lösungen entwickeln und auf den Markt bringen. Mehr entdecken: 👉 go.aws/480HUps