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.

``` mermaid 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 ``` ### 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. ```sh $ 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 -v $ pluginserver_python_query_cmd = /opt/kong-python-pdk/kong-pluginserver --no-lua-style --plugins-directory --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: raj713335@gmail.com

Gem.ai

Face Emotion & Age Detection System

AgentEval (AutoGen) Sample Implementation

Readme in japanese

Prompt Validator - Ensuring Safety with OpenAI Prompts

Hybrid Visual Impairment Assistance Tool Using Computer Vision

Sprinkler Controller

Computer Agent

Leveraging Machine Learning for Customer Lifetime Value Estimation

Creating AI Agents with Small Language Models

Sales Assistant Agent

Langfuse Kong Plugin

RSwarm AI agent library, with OpenAI’s API

nuwa-terminal-chat

Comprehensive Guide to AdaBoost: Theory and Applications

Image Segmentation Techniques In COMPUTER VISION

MediCarousel - Portable AI-powered Surgical Assistant

Quranic Text Analysis Through NLP

Weathercall-forecast-AT-test

Forecasting SPOT Price in the Spanish Electricity Market using Temporal Fusion Transformers