We're phasing out the N-to-1 retrieval strategy on September 1, 2024, and introducing a more flexible multi-path retrieval strategy. We recommend switching to this new approach to boost your application's retrieval efficiency.

Today, we're proud to announce that we've made DifySandbox open source for greater code transparency.

This blog thoroughly outlines the rationale, design principles, and implementation mechanisms that guided the development of DifySandbox.

With simple configuration, you can now access detailed application data, making it easier to evaluate the cost, latency, and quality of LLM applications created on Dify.

Dify v0.6.11 has just been released! We've integrated Firecrawl as our web data source solution, which greatly enhances the knowledge base of Dify.

Compare Dify.AI and LangChain for enterprise AI application development. Learn about their scalability, flexibility, and efficiency to choose the best platform for your needs.

Since launching workflow v0.6.0, it has been popular among our users, with over 90% needing it for AI applications in business scenarios. Dify v0.6.9 introduces publishing workflows as tools, iteration nodes, parameter extractors, and enhanced node capabilities. We also offer a showcase to demonstrate their use, automating batch processing of your emails.

We’re thrilled to launch our new feature: AI Workflow. It enables predictable outputs with multi-step logic through an intuitive, drag-and-drop interface.

Perfect for small teams & service providers needing custom branding & flexible deployment.

Explore how to leverage Google's Gemma, an open-source LLM, for integration with Dify. Discover tips for enhancing AI applications responsibly.

Ever wondered how to leverage the latest AI technology for travel planning? This comprehensive guide on Dify.AI showcases how to build an AI travel consultant, making your trip planning effortless and efficient.

Discover the step-by-step guide on constructing a sophisticated AI investment analysis assistant with Dify. Master the art of integrating AI into your financial analysis for smarter, data-driven decisions.

With Dify, you can create an AI Agent using any LLMs. Simply use instructions and combine various tools with additional knowledge bases to create a custom Assistant that's just right for you.

We've totally revamped Dify's core architecture, moving to a more modular approach. Our latest Beehive architecture allows each module to stand on its own, enabling developers to adjust parts without affecting the overall structure.

We're rolling out a brand new Model Runtime architecture, replacing the pre-existing model interface built on LangChain.

This article introduces how to use Dify.AI's Annotation Reply feature to improve Chatbot reply quality and Reduce LLM token Costs.

Dify.AI‘s v0.3.34 introduces features such as Annotation Reply, providing enhanced Q&A capabilities and more.

Jina AI's jina-embeddings-v2, with a unique 8,192 token context, enhances RAG applications in Dify.AI's platform. It outperforms standard 512-token models, enabling richer, context-aware AI solutions and simplified development.

OpenAI's Assistants API marks a shift in application engineering towards advanced AI use, emphasizing orchestration services. Dify, an open-source leader in this field, offers self-hosting for data security, multi-model support, and a flexible RAG engine. It enables privacy, compliance, customizable data processing, and team collaboration, enhancing AI application development and integration.

Dify.AI's latest update enhances its RAG technology, boosting QA efficiency in LLMs. It introduces hybrid search, a semantic rerank model, and multi-path retrieval, demonstrating significant performance improvements and a notable edge over OpenAI's Assistants API.

This article discusses enhancing RAG systems with Hybrid Search and Rerank technologies, focusing on improving retrieval accuracy and efficiency using LLMs for more comprehensive and precise search results.

This update officially supports the multimodal capabilities of the GPT4-Vision model, integrating images into conversations to create a unique interactive experience, making conversations more engaging.

v0.3.29 Feature Highlights: - GPT-4 Turbo & Vision model support - External data API tools - Moderation for sensitive content

Dify.AI introduces a new "Expert Mode" in its prompt orchestration page, offering advanced customization and control for professional developers and prompt engineers over prompt orchestration. While "Basic Mode" simplifies application creation, "Expert Mode" provides a wide range of tools and options for designing complex prompt orchestrations, enabling users to fine-tune AI applications for optimal performance. It includes features like customized prompts, different models (CHAT and COMPLETE) for various requirements, and a 'Log View' feature for debugging.

The recent updates to Dify.AI's dataset management tools introduce a "Citations and Attributions" feature for easier documentation referencing, and a new Dataset API for efficient data management. Support for multiple file formats and document segmentation enhances data handling. Additionally, integration with GPT-3.5-turbo-instruct and various Hugging Face embedding models provides users with more model options for specific applications, improving overall user experience.

Dify now supports popular open-sourced large language models like Llama2 and Qwen. Users can easily access these cutting-edge models by entering API keys on Dify, and build high-performance AI applications within minutes. Dify opens the gateway to the latest innovations in natural language processing. Try out top-notch models like Llama2 for free on Dify today, and unleash their power to enhance your AI solutions.

Embedding is a vital technique in AI applications, representing concepts as numerical sequences to enable better comprehension of relationships.

Dify.AI V0.3.12 introduces improved data retrieval with Q2Q matching and 'Chat' capability for online interactions with AI. Chat supports web browsing, Google search, and Wikipedia queries, enhancing productivity. Join our collaborative exploration of large model capabilities.

Dify V0.3.9 Release: Integration of Antropic's Claude2 and Claude-instant models, supporting ultra-long conversations with up to 100K tokens. Now, you can embed AI applications on your web pages, swiftly creating personalized AI customer support for your official website. Additional features include bulk result exporting for text-based applications and the community version supporting member invitations through invitation links. Moreover, the Service API now includes conversation deletion functionality. Experience the superpowers of Dify today!

Since the official launch of the Cloud version on May 9th, Dify.AI has been well-received and widely shared by developers. In less than a week, over 4,000 applications have been created, and even before we went open source, our GitHub Star count had exceeded 700+. This makes us deeply feel the power of the community! At the same time, we are extremely honored and excited that Dify can bring such powerful creativity to developers.

With the emergence of various capabilities of large language models (LLMs), the scenarios for AI applications have become much broader. However, for most developers, developing AI applications based on large language models such as GPT and technology frameworks like Langchain is still a challenging task. Developers must spend a lot of time learning various obscure concepts and technical research, and they cannot carry out continuous operations of AI applications.

Developers utilizing OpenAI's GPT API may often find that the AI doesn't produce the desired output or seems "uncooperative." To address these challenges and improve GPT's performance, there are three crucial steps to follow：

LLMOps is a specialized area within the broader MLOps landscape, focusing on the deployment, management, and improvement of AI applications built on top of Large Language Models (LLMs) like GPT-4. In this article, we will explore the concept of "Ops" in LLMOps and how Dify caters to this space.

At Dify, we are dedicated to providing our users with seamless and flexible solutions for creating and deploying AI-native applications. We understand the importance of a user-friendly and open platform to foster innovation, and that's where Dify WebApp and Dify WebApp Template come into play.

At Dify, we are passionate about making AI technology more accessible and user-friendly. As we began our journey, we identified several key challenges that developers and users face when working with AI-native applications, especially with the advanced large language models provided by OpenAI. In this blog post, we will discuss the reasons behind the creation of Dify and our mission to simplify the AI development process.

The combination of LLM embeddings and datasets has dramatically transformed the MLOps landscape, unlocking new capabilities and driving innovation in AI applications. Dify's dataset functionality simplifies the process of integrating proprietary data with LLMs, empowering developers to build more intelligent, domain-specific AI solutions. As LLMs continue to evolve, we can expect even more exciting possibilities and advancements in the world of AI and MLOps.

ChatGPT plugins provide a powerful way to extend the capabilities of AI clients by integrating additional functionalities through APIs. In this blog post, we'll guide you through the process of creating a ChatGPT plugin and discuss some potential business use cases. Finally, we'll touch upon how Dify supports the integration of both in-house and third-party plugins for seamless AI application orchestration.