Extract PDF text in your browser with LiteParse for the web

Community Bilge Yücel DevRel Engineer Haystack Ecosystem: One Name, One Product Family, One Look One unified Haystack ecosystem, from open source to enterprise-scale AI systems. December 19, 2025

Haystack Ecosystem: One Name, One Product Family, One Look One unified Haystack ecosystem, from open source to enterprise-scale AI systems. December 19, 2025We’re making some naming and visual updates at deepset to better reflect the role Haystack already plays as a framework, a community, and the foundation of our enterprise platform. If you’re already building with Haystack, nothing is changing in how you build or run applications. This update is about clarity, making the Haystack ecosystem easier to understand, easier to navigate, and centered around a single open foundation. The Open Source to Enterprise Story of Haystack Haystack began as an open-source framework for building NLP pipelines, created to give developers precise control over how AI systems are composed, debugged, and run in production. From the start, it was designed for real-world use, not just experimentation. Over time, the framework…

Community Bilge Yücel DevRel Engineer Announcing Haystack Enterprise Starter: Best Practices and Support A Faster Way to Build and Scale Production-Grade AI Apps August 1, 2025

Announcing Haystack Enterprise Starter: Best Practices and Support A Faster Way to Build and Scale Production-Grade AI Apps August 1, 2025💙 Thanks to you and all of our amazing community members, the Haystack open source framework has grown into a thriving developer ecosystem, now used by thousands of organizations to power everything from simple Q&A bots to advanced enterprise agents. As more teams run Haystack in production, one thing has become increasingly clear: building reliable AI systems is hard and scaling them securely is even harder. We’ve had a front-row seat to these challenges. Across GitHub threads, meetups, community calls, and production deployments, developers have consistently asked for engineering support and hands-on guidance to build for their use case, accelerate deployment, improve observability, and scale infrastructure with confidence. These aren’t just feature requests; they reflect the real-world friction points of…

Next

Blog Articles about Haystack, LLMs, Agents, and latest AI technologies. All articles Use DeepSeek-R1 with Haystack: Demo and Tutorial Compare DeepSeek-R1 and OpenAI's o1 in the deepset demo and explore their reasoning capabilities January 29, 2025Build an Agentic RAG Pipeline in deepset Studio Use deepset Studio to build an agentic Haystack pipeline with a fallback mechanism for dynamic web search January 14, 2025Announcing Advent of Haystack 2024 🎄 Join the Festive AI Fun! December 2, 2024Create a Swarm of Agents Easy creation of multi-agent systems November 26, 2024Announcing Studio: Your Development Environment for Haystack Build, deploy, and test Haystack pipelines with ease November 20, 2024Building a Multimodal Nutrition Agent Use fastRAG and Haystack to build an agent that can process text and image data November 7, 2024Design Haystack AI Applications Visually in deepset Studio with NVIDIA NIM November 1, 2024Advanced…

4M Models Scanned: Protect AI + Hugging Face 6 Months In

4M Models Scanned: Protect AI + Hugging Face 6 Months In Hugging Face and Protect AI partnered in October 2024 to enhance machine learning (ML) model security through Guardian’s scanning technology for the community of developers who explore and use models from the Hugging Face Hub. The partnership has been a natural fit from the start—Hugging Face is on a mission to democratize the use of open source AI, with a commitment to safety and security; and Protect AI is building the guardrails to make open source models safe for all. 4 new threat detection modules launched Since October, Protect AI has significantly expanded Guardian's detection capabilities, improving existing threat detection capabilities and launching four new detection modules: - PAIT-ARV-100: Archive slip can write to file system at load time - PAIT-JOBLIB-101: Joblib model suspicious code execution detected at model…

Visual Document Retrieval Goes Multilingual

Visual Document Retrieval Goes Multilingual TL;DR: We presentvdr-2b-multi-v1 , the best multilingual embedding model for visual document retrieval. We also release its English-only twinvdr-2b-v1 and open-source the newvdr-multilingual-train dataset. With 500k high-quality samples, it's the largest open-source multilingual synthetic dataset for visual document retrieval. Introducing vdr-2b-multi-v1 (🤗), a multilingual embedding model designed for visual document retrieval across multiple languages and domains. This model is designed to encode document page screenshots into dense single-vector representations, this will effectively allow to search and query visually rich multilingual documents without the need for any OCR, data extraction pipelines, chunking... The vdr-2b-multi-v1 model is based on MrLight/dse-qwen2-2b-mrl-v1 and is trained on an extensive self-made dataset of multilingual query-image pairs. This model is built in collaboration with LlamaIndex and is the next iteration of mcdse-2b-v1 . Our vdr-2b-multi-v1 extends and improves the learning and methods…

Hugging Face x LangChain : A new partner package

Hugging Face x LangChain : A new partner package in LangChain langchain_huggingface , a partner package in LangChain jointly maintained by Hugging Face and LangChain. This new Python package is designed to bring the power of the latest development of Hugging Face into LangChain and keep it up to date. From the community, for the community All Hugging Face-related classes in LangChain were coded by the community, and while we thrived on this, over time, some of them became deprecated because of the lack of an insider’s perspective. By becoming a partner package, we aim to reduce the time it takes to bring new features available in the Hugging Face ecosystem to LangChain's users. langchain-huggingface integrates seamlessly with LangChain, providing an efficient and effective way to utilize Hugging Face models within the LangChain ecosystem. This partnership is not just about…

Introducing the Chatbot Guardrails Arena

Introducing the Chatbot Guardrails Arena Lighthouz AI is therefore launching the Chatbot Guardrails Arena in collaboration with Hugging Face, to stress test LLMs and privacy guardrails in leaking sensitive data. Put on your creative caps! Chat with two anonymous LLMs with guardrails and try to trick them into revealing sensitive financial information. Cast your vote for the model that demonstrates greater privacy. The votes will be compiled into a leaderboard showcasing the LLMs and guardrails rated highest by the community for their privacy. Our vision behind the Chatbot Guardrails Arena is to establish the trusted benchmark for AI chatbot security, privacy, and guardrails. With a large-scale blind stress test by the community, this arena will offer an unbiased and practical assessment of the reliability of current privacy guardrails. Why Stress Test Privacy Guardrails? Data privacy is crucial even if you…

Open-source LLMs as LangChain Agents

Open-source LLMs as LangChain Agents TL;DR Open-source LLMs have now reached a performance level that makes them suitable reasoning engines for powering agent workflows: Mixtral even surpasses GPT-3.5 on our benchmark, and its performance could easily be further enhanced with fine-tuning. We've released the simplest agentic library out there: smolagents! Go checkout the smolagents introduction blog here. Introduction Large Language Models (LLMs) trained for causal language modeling can tackle a wide range of tasks, but they often struggle with basic tasks like logic, calculation, and search. The worst scenario is when they perform poorly in a domain, such as math, yet still attempt to handle all the calculations themselves. To overcome this weakness, amongst other approaches, one can integrate the LLM into a system where it can call tools: such a system is called an LLM agent. In this post,…

langfuse ClickHouse Langfuse joins ClickHouse Our goal continues to be building the best LLM engineering platform Read story

Langfuse joins ClickHouse Our goal continues to be building the best LLM engineering platform ClickHouse has acquired Langfuse. If you're reading this as a Langfuse user, your first question is probably: What does this mean for me? Our roadmap stays the same, our goal continues to be building the best LLM engineering platform, and we remain committed to open source and self-hosting. There are no immediate changes to how you use Langfuse and how you can reach out to us. What does change is our ability to move faster. With ClickHouse behind us, we can invest more deeply into performance, reliability, and our roadmap that helps teams build and improve AI applications in production. What stays the same This is the section we would want to read first, too. - Langfuse stays open source and self‑hostable. There are no planned…

Object Detection for Dummies Part 2: CNN, DPM and Overfeat

Part 1 of the “Object Detection for Dummies” series introduced: (1) the concept of image gradient vector and how HOG algorithm summarizes the information across all the gradient vectors in one image; (2) how the image segmentation algorithm works to detect regions that potentially contain objects; (3) how the Selective Search algorithm refines the outcomes of image segmentation for better region proposal. In Part 2, we are about to find out more on the classic convolution neural network architectures for image classification. They lay the foundation for further progress on the deep learning models for object detection. Go check Part 3 if you want to learn more on R-CNN and related models. Links to all the posts in the series: [Part 1 ] [Part 2 ] [Part 3 ] [Part 4 ]. CNN for Image Classification# CNN, short for “Convolutional…

Building LLM-Powered Web Apps with Client-Side Technology October 13, 2023 Recreate one of the most popular LangChain use-cases with open source, locally running software - a chain that performs Retrieval-Augmented Generation, or RAG for short, and allows you to “chat with your documents”

Building LLM-Powered Web Apps with Client-Side Technology October 13, 2023 This is a guest blog post by Jacob Lee, JS/TS maintainer at @LangChainAI, formerly co-founder & CTO at @Autocode, engineer on Google photos. The initial version of this blog post was a talk for Google’s internal WebML Summit 2023, which you can check out here: It’s no secret that for a long time machine learning has been mostly a Python game, but the recent surge in popularity of ChatGPT has brought many new developers into the field. With JavaScript being the most widely-used programming language, it’s no surprise that this has included many web developers, who have naturally tried to build web apps. There’s been a ton of ink spilled on building with LLMs via API calls to the likes of OpenAI, Anthropic, Google, and others, so I thought I’d…

The Sora feed philosophy

Our aim with the Sora feed is simple: help people learn what’s possible, and inspire them to create. Here are some of core starting principles to bring this vision to life: - Optimize for creativity. We’re designing ranking to favor creativity and active participation, not passive scrolling. We think this is what makes Sora joyful to use. - Put users in control. The feed ships with steerable ranking, so you can tell the algorithm exactly what you’re in the mood for. Parents can also turn off feed personalization and control continuous scroll for their teens through ChatGPT parental controls. - Prioritize connection. We want Sora to help people strengthen and form new connections, especially through fun, magical Cameo flows. Connected content will be favored over global, unconnected content. - Balance safety and freedom. The feed is designed to be widely…

Updating our Model Spec with teen protections

We’re sharing an update to our Model Spec, the written set of rules, values, and behavioral expectations that guides how we want our AI models to behave, especially in difficult or high stakes situations, with Under-18 (U18) Principles(opens in a new window). Model behavior is critical to how people interact with AI, and teens have different developmental needs than adults. The U18 Principles guide how ChatGPT should provide a safe, age-appropriate experience for teens aged 13 to 17. Grounded in developmental science, this approach prioritizes prevention, transparency, and early intervention. In developing these principles, we previewed them with external experts, including the American Psychological Association, as part of our ongoing work to seek input to strengthen our approach. While the principles of the Model Spec continue to apply to both adult and teen users, this update clarifies how it should…

OpenAI’s approach to AI and national security

OpenAI’s approach to AI and national security Today, the White House released a National Security Memorandum (NSM) on Artificial Intelligence(opens in a new window) outlining how the U.S. government can responsibly harness AI to advance national security while establishing essential guardrails for its use. The NSM also recognizes the importance of increasing the supply and access to semiconductor chips, power generation, and data center capacity – all of which we agree are essential to continued U.S. leadership on AI. At OpenAI, we’re building AI to benefit the most people possible. Supporting U.S. and allied efforts to advance AI in a way that upholds democratic values is essential to our mission of ensuring AI’s benefits are widely shared. We view the NSM as an important step forward in that effort – here is how we’re currently thinking about national security and…

DALL·E 2 pre-training mitigations

DALL·E 2 pre-training mitigations In order to share the magic of DALL·E 2 with a broad audience, we needed to reduce the risks associated with powerful image generation models. To this end, we put various guardrails(opens in a new window) in place to prevent generated images from violating our content policy(opens in a new window). This post focuses on pre-training mitigations, a subset of these guardrails which directly modify the data that DALL·E 2 learns from. In particular, DALL·E 2 is trained on hundreds of millions of captioned images from the internet, and we remove and reweight some of these images to change what the model learns. This post is organized in three sections, each describing a different pre-training mitigation: - In the first section, we describe how we filtered out violent and sexual images from DALL·E 2’s training dataset.…

Learning to summarize with human feedback

Learning to summarize with human feedback We’ve applied reinforcement learning from human feedback to train language models that are better at summarization. Why it matters Our models generate summaries that are better than summaries from 10x larger models trained only with supervised learning. Even though we train our models on the Reddit TL;DR dataset, the same models transfer to generate good summaries of CNN/DailyMail news articles without any further fine-tuning. Our techniques are not specific to summarization; in the long run, our goal is to make aligning AI systems with human preferences a central component of AI research and deployment in many domains. Large-scale language models are becoming increasingly capable on NLP tasks. These models are usually trained with the objective of next word prediction on a dataset of human-written text. But this objective doesn’t capture exactly what we want;…

PixelCNN++: Improving the PixelCNN with discretized logistic mixture likelihood and other modifications

PixelCNN++: Improving the PixelCNN with discretized logistic mixture likelihood and other modifications Abstract PixelCNNs are a recently proposed class of powerful generative models with tractable likelihood. Here we discuss our implementation of PixelCNNs which we make available at this https URL(opens in a new window). Our implementation contains a number of modifications to the original model that both simplify its structure and improve its performance. 1) We use a discretized logistic mixture likelihood on the pixels, rather than a 256-way softmax, which we find to speed up training. 2) We condition on whole pixels, rather than R/G/B sub-pixels, simplifying the model structure. 3) We use downsampling to efficiently capture structure at multiple resolutions. 4) We introduce additional short-cut connections to further speed up optimization. 5) We regularize the model using dropout. Finally, we present state-of-the-art log likelihood results on CIFAR-10…