A Java framework for LLM applications
lightweight, simple, and more elegant.
Agents-Flex supports common large language models on the market, supports the Ollama deployment model, and also has built-in rich network protocols for docking large models, such as HTTP, SSE, WS, etc., so that developers can use it to easily dock various large models.
Agents-Flex provides a wealth of large language model development templates and support for Prompt Framework, such as FEW-SHOT, CRISPE, BROKE, ICIO, etc. In addition, Prompt Template we can also customize our own unique content.
Agents-Flex has a very flexible Function Calling component. It supports local method definitions, parsing, callbacks through LLMs, and executing local methods to obtain results.
Agents-Flex offers Loader, Parser, and Splitter components for the Document. Each component has multiple implementations, making it easy to load data from the web, local files, databases, and various data types.
The Memory module of Agents-Flex is divided into MessageMemory and ContextMemory, used for recording chat messages and Chain execution contexts. Developers can extend the Memory module by inheritance to achieve richer functionalities.
Agents-Flex includes extensive embedding capabilities and extensions. Developers can implement the Embedding interface to expand their private embedding algorithms and support.
Agents-Flex supports multiple vector databases. Developers can also implement the VectorStore interface to expand their private VectorStore services.
Agents-Flex’s Chain includes sequential Chains, asynchronous Chains, and loop Chains, helping developers handle various scenarios.