Langfuse-Kong-Plugin

Langfuse is an Open Source LLM Engineering Platform to log Traces, evals, prompt management, and metrics to debug and improve your LLM application.

Read Langfuse documentation.

Key Features of Langfuse:

• Logging and Tracing: Langfuse captures detailed logs of interactions with LLMs, allowing developers to trace the exact steps taken during a conversation or process. This helps identify where things might have gone wrong or improvements can be made.

• Visualization: The tool provides visual representations of data flows, decision trees, and other critical aspects of the LLM's operations. This can be particularly useful in complex applications where understanding the flow of information is crucial.

• Debugging: By offering granular insights into the operations of LLMs, Langfuse aids in debugging issues that might not be immediately apparent. This includes tracking down errors, understanding unexpected behavior, and optimizing responses.

• Performance Monitoring: Langfuse helps in monitoring the performance of LLM applications over time, identifying trends, and detecting anomalies. This is essential for maintaining the efficiency and reliability of applications that rely on LLMs.

• Integrations: Langfuse often supports integrations with other tools and platforms commonly used in the development and deployment of LLMs, making it easier to incorporate into existing workflows.

1. Sequence Diagram

Langfuse Kong Plugin Flow to add Logs of the GenAI LLM usage directly from Gateway to monitor usage and cost.

sequenceDiagram

    Title: Langfuse Kong Plugin Flow to add Logs of the GenAI LLM usage directly from Gateway to monitor usage and cost.  
    
    participant Client API
    participant Kong API Gateway
    participant Langfuse-Kong-Plugin
    participant LLM Endpoint
    participant Langfuse Dashboard
    
    Client API->>Kong API Gateway: API call redirected to Kong API Gateway 
    
    Kong API Gateway->>Langfuse-Kong-Plugin: API gateway intercepts the call and trigger the plugin  
    
    
    Langfuse-Kong-Plugin->>LLM Endpoint: Plugin redirects the call to the configured LLM Provider and fetches the generated response
    
    Note right of Langfuse-Kong-Plugin: Can be configured to use any LLM Provider

    
    LLM Endpoint-->>Langfuse-Kong-Plugin: LLM provider sends the generated response
    
    Langfuse-Kong-Plugin->>Langfuse Dashboard: Plugin add the log traces to the Langfuse Dashboard
    
    Langfuse-Kong-Plugin-->>Kong API Gateway: Plugin sends the generated response
    
    Kong API Gateway-->>Client API: API Gateway sends the generated response

Loading

Tested in Kong Release

• Kong Community 3.7.1

Setup Up Kong Locally Follow the GitHub Repository

2. Getting Started With the Langfuse Application (Self-host docker)

• Clone this repository and open it up in your IDE or terminal
• Run cp .env.example .env and replace values appropriately in the .env file
• Run docker compose up -d to get the langfuse instance running.

$ cd Langfuse-Kong-Plugin
$ docker-compose up -d --build

• Navigate to http://localhost:3000
• Click Sign up to create your account

Dashboard

• Create a new project
• Go to settings and click Create new API keys
• Copy the Secret Key and Public Key
• Store keys in the .env file within your LLM project:

LANGFUSE_SECRET_KEY="sk-your-key"
LANGFUSE_PUBLIC_KEY="pk-your-key"
LANGFUSE_HOST="http://localhost:3000"

3. Langfuse-Kong-Plugin Plugin Installation

Install Kong Pdk and Python Packages

$ apt-get update && apt-get add python3 py3-pip python3-dev musl-dev libffi-dev gcc g++ file make && PYTHONWARNINGS=ignore pip3 install kong-pdk
$ pip install -r requirements.txt

Make the Below Changes in Kong.conf

$ pluginserver_names=python
$ pluginserver_python_socket=/usr/local/kong/python_pluginserver.sock
$ pluginserver_python_start_cmd = /opt/kong-python-pdk/kong-pluginserver --no-lua-style --plugins-directory <PATH_OF_PLUGIN_FOLDER> -v
$ pluginserver_python_query_cmd = /opt/kong-python-pdk/kong-pluginserver --no-lua-style --plugins-directory <PATH_OF_PLUGIN_FOLDER> --dump-all-plugins

After Installing the Plugin using any of the above steps . Add the Plugin Name in Kong.conf

plugins = bundled,Langfuse-Kong-Plugin

Restart Kong

kong restart

4. Configuration Reference

Enable the plugin on a Route

Admin-API

For example, configure this plugin on a service by making the following request:

curl -X POST http://{HOST}:8001/routes/{ROUTE}/plugins \
--data "name=Langfuse-Kong-Plugin"  \
--data "config.OPENAI_API_KEY={OPENAI_API_KEY}"
--data "config.OPENAI_API_MODEL={OPENAI_API_MODEL}"
--data "config.LANGFUSE_SECRET_KEY={LANGFUSE_SECRET_KEY}"
--data "config.LANGFUSE_PUBLIC_KEY={LANGFUSE_PUBLIC_KEY}"
--data "config.LANGFUSE_HOST={LANGFUSE_HOST}"

Declarative(YAML)

For example, configure this plugin on a service by adding this section to your declarative configuration file:

routes : 
    name: {ROUTE}
    plugins:
    - name: Langfuse-Kong-Plugin
    config:
        OPENAI_API_KEY: {OPENAI_API_KEY}
        OPENAI_API_MODEL : {OPENAI_API_MODEL}
        LANGFUSE_SECRET_KEY : {LANGFUSE_SECRET_KEY}
        LANGFUSE_PUBLIC_KEY : {LANGFUSE_PUBLIC_KEY}
        LANGFUSE_HOST : {LANGFUSE_HOST}
    enabled: true
    protocols:
    - grpc
    - grpcs
    - http
    - https

ROUTE is the id or name of the route that this plugin configuration will target. OPENAI_API_KEY is the openAI Key . MODEL_NAME is the AI Model name LANGFUSE_SECRET_KEY is the Langfuse Secret Key. LANGFUSE_PUBLIC_KEY is the Langfuse Public Key. LANGFUSE_HOST is the Langfuse Host Url.

Parameters

FORM PARAMETER	Type	DESCRIPTION
ROUTE		The name of the Route the plugin targets.
config.OPENAI_API_KEY	Type:string	OpenAI Secret Key
config.OPENAI_API_MODEL	Type:string	OpenAI Model Name
config.LANGFUSE_SECRET_KEY	Type:string	LANGFUSE SECRET KEY
config.LANGFUSE_PUBLIC_KEY	Type:string	LANGFUSE PUBLIC KEY
config.LANGFUSE_HOST	Type:string	LANGFUSE HOST URL

5. Project Requirements

Languages and Tools version used

• Python 3.11.4
• Langfuse 2.73.0
• Kong 3.7.1

6. Application Screenshots / Demo.

Contributors

Design & Developed By: [email protected]