Bilge

🚀 Haystack 2.26 is out! Here are a few highlights: 🤖 More flexible Agents with dynamic system prompts You can now use Jinja2 templating in `system_prompt` to inject runtime variables and dynamically adapt behavior, such as changing language or tone, or adding conditional logic, without redefining prompts. 🔍 LLMRanker for smarter reranking LLMRanker uses an LLM to reason about relevance rather than rely on similarity scores (like cross-encoders). This leads to better results for complex queries and helps provide higher-quality context for RAG and agent workflows while keeping context windows lean. 🧠 Discover supported models programmatically Chat generators now expose SUPPORTED_MODELS variable, so you can easily see which models are supported. 💙 Big thanks to our contributors to this release!

@infornomics @krotenWanderung Thanks! it looks like it’s gonna be sunny all day ☀️

@bilgeycl @krotenWanderung Crossing my fingers for you that at least the weather is great

Sooo there’s a transport strike in Munich tomorrow… (thx @krotenWanderung for jinxing it 🙃) But the unconference is still ON 💪 Perfect excuse to bike in the sun 🚲 or, if you’re a public transport fan like me, some buses & S-Bahns are still running, check them here 👇 #trip@origdest" target="_blank" rel="nofollow noopener">efa.mvv-muenchen.de/index.html?nam… See you there 🍻

@krotenWanderung lol I love it 😂

@bilgeycl Onconference (I am just good with probabilities estimation…😒)

🕒 We run them every 2 weeks on Tuesday at 3PM CET | 9AM ET | 6:30PM IST 👇 luma.com/5st62yw2

Last Office Hours we talked about: - using coding agents (Claude Code) to bootstrap @Haystack_AI apps - evaluation-driven pipeline iteration + @MLflow judges - skills for Haystack (shoutout to @srini047 🙌) Lately, people don’t really come to ask questions. They come to listen, get inspired, and see how others are building. And honestly, I learn a lot from these sessions too We're hosting the next one on March 24, come join & tell @LukawskiKacper and me what you’re building 👇

@qdrant_engine yesss can’t wait!

⏳ Just a few days left! 🇩🇪 Join us in Munich for “Unconference: Context Engineering with Open-Source” - an evening of open discussions on building better context layers for AI systems. Instead of traditional talks, we’ll dive into interactive conversations around: • Retrieval and vector search • Memory systems for AI agents • Designing robust context layers • Building production-grade AI with open-source tools Explore how tools like @deepset_ai, @qdrant_engine, and @cognee_ help engineers design better context pipelines for real-world AI applications. 📍 smartvillage Bogenhausen, Munich 🔗 luma.com/xg649r8v See you there! 👋

Just published a notebook on @GeminiApp Cookbook w/ @bilgeycl 📓 Cross-modal Retrieval with Gemini Embedding 2 and @Haystack_AI: github.com/google-gemini/… 🔎 Several examples of search: text→text text→any (PDF included) image→text search Thx @patloeber! @googleaidevs

@malgamves Noir. I’m not a Parisienne, but I worked at this cafe and really enjoyed it. Great interior👌🏼

what cafes in paris let you sit on your laptop and write your substack on weekends?

👩‍🎤 Our Call for Papers is open! Submit your talk and workshop proposals by April 10 💾 For more details visit divaconf.com/cfp 📝 #diva2026 #diveintoAI #cfp #tech #AIconference

@mervenoyann Mayısta hazır düzenlenmekte olan bir yapay zeka konferansı var: @divaconference 💜 Ben orada olucam, hf de gelse ne güzel olur! Sponsorluk çağrısı hala açık mı @apatheticmagpie?

istanbul'da olan arkadaşlarım 👋🏻 eğer açık kaynak yapay zeka üstüne meet-up, hackathon vs düzenleme niyetindeyseniz sizi desteklemeyi çok isteriz 🤗 başvurmak için bana da yazabilirsiniz ⤵️

📚 It had been a while since we published a technical blog post, and Gemini Embedding 2 from @GoogleDeepMind team felt like a good reason to write one together with @theanakin87 💙 It's Google's first fully multimodal embedding model, meaning you can embed text, images, audio, video, and PDFs into a single shared vector space. What’s exciting about multimodal embedding models is that they enable cross-modal retrieval. For example, you can search a product catalog with a text query like “grey running shoes with white soles” 👟, or upload a photo of a monument to retrieve relevant articles describing it 🗽 In the post, we show how to use Gemini Embedding 2 with @Haystack_AI for both text and multimodal data, supported from day 0 🎉 Blog: haystack.deepset.ai/blog/multimoda…

@mervenoyann @malgamves @swyx @aiDotEngineer It’s gonna be a great conference!

I'm big fan of @swyx @aiDotEngineer youtube channel (hands-down my favorite) so stoked to be accepted to the London one see y'all there!

🚀 @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…

Three cities. Two countries. One AI event organized by four women working in AI 💜 Around International Women’s Day, it feels especially meaningful to share this. Together with @HandexxK, @krotenWanderung, Paula Plattner, and yours truly, we’re organizing the next edition of our unconference in Munich: Context Engineering with Open-Source Instead of slides and talks, we’re creating a space for real conversations about the context layer of AI systems — retrieval, memory, orchestration — and what it takes to run AI reliably in production. Expect roundtable discussions with engineers, product leaders, and decision-makers who are actively building and operating AI systems. We’re excited to bring together people from across the open AI ecosystem, including teams behind @deepset_ai/@Haystack_AI , @qdrant_engine, and @cognee_ If you're building AI systems, evaluating tools, or thinking about how to design reliable AI infrastructure — this event is for you. 📍 smartvillage Bogenhausen, Munich 📅 March 19 🕕 18:00 CET 👉 Join us: luma.com/xg649r8v

🚀 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 💬

but i need a company to fund my token usage 😫

We had a long conversation about AI in the public sector, a space I’m trying to understand better. It’s easy to talk about LLMs and agents… harder when you have to integrate with 5 different legacy systems in a compliant way

a bit short notice 😅 but @LukawskiKacper and I are hosting an Office Hour today at 3PM on the @Haystack_AI Discord Bring your questions about Haystack, agents, RAG or your experiments. You might also get a sneak peek of the upcoming release 👀 Join us: luma.com/ghs6tz86

I’ll be speaking at @pyconit this May! 🇮🇹🐍 My talk will focus on practical agentic system design: why most AI agents fail in real-world settings, how to structure the right context and interactions, and how to build systems that are modular, observable, and maintainable. I’ve already spotted some familiar faces in the speaker list 👀 Looking forward to reconnecting with old friends, meeting new ones, and spending time with the amazing Python & @Haystack_AI community in Italy 🍝 If you’re attending, let’s connect! ✨