News from the Dify team! The latest version introduces features such as Annotation Reply, providing enhanced Q&A capabilities and more.

New Features

Annotation Reply

Compared to earlier times, when we could only passively accept the Reply of LLMs, we now have more control. The Annotation Reply feature delivers customized, high-quality responses for different scenarios via manual annotation.

These manually modified annotations are stored for future reference, enabling quick access for similar queries, thus skipping the typical generation phase of LLMs and saving on token usage. The saved annotation data can be adjusted and exported, crucial for LLMOps. It maintains the ongoing operation of apps and provides data for future model fine-tuning. >>Learn More

Unstructured.io Support

Unstructured seamlessly extracts and transforms intricate data, making it compatible with vector databases and LLM frameworks. It simplifies the process, ensuring the data is easy to handle and clean.

We're excited to announce the integration with unstructured.io as the preferred solution for ETL (Extract, Transform, Load) processes in our Cloud Service. This update adds support for four new text parsing formats (msg, eml, ppt, xml) and optimizes two existing ones (text, markdown).

You can now easily upload this unstructured data without any extra processing steps.

New Models

Support is available for Azure OpenAI's GPT-4-1106-preview and GPT-4-vision-preview models. The GPT-4-1106-preview model excels in delivering quicker responses and is more budget-friendly. On the other hand, the GPT-4-vision-preview model stands out for its superb multimodal capabilities. Special thanks to community developer @charli117 for their crucial contributions.

Customizable SaaS Services

In our latest SaaS version, subscribers can now customize the Web App by replacing the default logo, in addition to using team collaboration and more Vector Storage and Annotation Quota, to better meet your customization needs.

Notice

In selfhost, in order for users to take advantage of the Annotation Reply feature, a required upgrade process entails executing a command within the API Docker container. Comprehensive instructions are available for both Docker and source code deployment methods. >>Learn More

Contributors

Welcome new contributors @guchenhe and @ethuwlwfu3288 to our community.

Further Reading

Explore the GitHub release notes and detailed Annotation Reply documentation for more comprehensive information.