Overview
Welcome to the LinguFlow User Guide! LinguFlow is a cutting-edge LLM (Large Language Model) application builder that facilitates the creation of LLM applications through a visual, DAG(Directed Acyclic Graph)-based interface.
This guide is designed to familiarize you with the capabilities of LinguFlow and navigate you through its wide array of features and functionalities.
After reading, feel free to dive into the QuickStart section to get hands-on experience with LinguFlow.
Key Features
Development
- LinguFlow UI: After deployment (locally or self-hosted), you can manage your LLM applications within an organized framework of Apps and Versions.
- App: Each LLM application is recognized as an
application
, which can be executed directly using itsapplication_id
during production. - Version: Each App can house multiple versions, with one specified as the
published version
. Each version can be independently edited.
- App: Each LLM application is recognized as an
- Builder: Opening a version reveals the Builder, where application logic can be crafted using a DAG (Directed Acyclic Graph) methodology.
- Block: Blocks serve as the nodes within the DAG, each embodying a distinct processing logic. Block types encompass input, output, LLM, third-party tools, numerical processing, and the invocation of other LinguFlow applications.
- Line: Lines link Blocks within the DAG, differentiated into data lines and decision lines (True/False).
- Debugging: The Debugging feature enables swift testing of a version post-application construction.
Run
- Call an application: Execute LinguFlow applications asynchronously through API calls, leveraging the
published version
of an application. - Tracing: LinguFlow supports LangFuse Cloud as its Tracing instrument, tracing every application interaction.