encoder-run: Kubernetes Operator for automatically generating vector embeddings for source code

encoder-run is a Kubernetes operator designed to automate the lifecycle of source code embeddings. It also manages the underlying storage and model infrastructure. This project ensures vectors stay current with the latest changes in source code repositories, enhancing search, similarity checks, RAG, and code analysis tasks.

Features

• Model Deployment: Deploys SentenceTransformer/Encoding models to process embeddings internally.
• Continuous Synchronization: Monitors source code repositories and triggers the encoding process for updated code, ensuring that embeddings remain current.
• Scalable and Efficient: Optimized for performance and scalability, handling multiple repositories and large codebases with ease.
• Customizable Encoding Models: Supports a variety of encoding models, allowing users to choose the best fit for their specific requirements.

Encoder Pipeline Architecture

Getting Started

Notes

• Ensure you have plenty of available memory to run the kubernetes setup. Check Docker Desktop resources for memory and make sure its 24GB+. Ensure disk is set reasonably as well due to the nature of large model weights and large docker images.
• Dimensions for embeddings is currently hard-coded to 768 so ensure the transformer you're deploying matches.
• Local embedding time is much longer than using a GPU based setup. Sometimes the models first request/response batch is slow but subsequent requests will be faster and more consistent. Feel free to run this setup on a machine with GPU access to speed up inference time.

Prerequisites

Before setting up encoder-run, ensure the following prerequisites are met:

• Golang: Goland version 1.21.*+ is required.
• Docker: Docker must be installed on your system as it is required to containerize the application components and dependencies. Visit Docker's official site for installation instructions.
• Kubernetes Kind: Kind is used to create a local Kubernetes cluster. If it is not already installed, you can run make kind-install which will check for its presence and install it if it's missing.
• Node: Node is used only if you are developing locally. Please install v18.17.0.

Deployment

After satisfying the prerequisites, use the following command to deploy the application. This will build all images from source. First time can take 10 minutes+.

make deploy

or to skip building from source (faster as it pulls from dockerhub):

make deploy-staging

This command will set up the necessary Kubernetes configurations and launch all components. Once deployed, it will expose:

• Console-UI on localhost:32081
• Gateway on localhost:32080

Running the Frontend with Mock Data

To run the frontend interface with mock data, follow these steps:

1. Navigate to the frontend console-UI directory:

   cd frontend/console-ui
   npm install && npm run generate
   cd ../mock
   npm install
   cd ../console-ui
   npm run local

This setup will spin up the UI on localhost:3000 and the mock gateway on localhost:4000, simulating the full environment for development or testing.

Roadmap

Here's a short list of planned features and enhancements for encoder-run:

• Redis deployments
• Postgres external db
• Postgres deployments
• ElasticSearch deployments/external
• Custom chunking strategies based on AST, etc

Feel free to contribute or suggest new features by opening an issue or submitting a pull request on our GitHub repository.

Community and Feedback

Encoder-run is an open-source project and we encourage and welcome contributions. If you wish to contribute, be sure to review our contribution guidelines and code of conduct.

For problems with the installation and setup, discussions about how to best to use encoder-run please use our Github issues and make sure to include as much detail as possible.