GoLLM (pronounced "Golem")

A simple HTTP Server & CLI-based TUI Application that interfaces with the OpenAI API in Golang.

Set-Up

Required for Both Modalities

• Open root directory of project
• Create .env file from .env.template file, providing OpenAI API Key

CLI-based TUI Application

• Run command: go run main.go cli
• This will initialize the application's TUI and allow you to begin prompting the LLM immediately with response rendered in the TUI (Memory is enabled/Memory Management needs enhancement)
  • CAUTION: Context Overflow is possible with the existing memory implementation

Gin-Gonic HTTP Server

• Run command: go run main.go or build binary (go build and/or go install) which will generate an executable named gollm
• Using a CLI-based tool like curl or a HTTP server testing GUI like Postman, create a new POST request to http://localhost:8080 (default port)
• In the body, create a JSON with a "text" key and assign your prompt as a value (in the form of a string)
• Initiate the request (Press return in the CLI or Press "Send" in Postman)
• View the Request's Response Payload for Assistant's Completion for the prompt sent

Primary Packages Used

• gin-gonic: Used to run the HTTP server and route incoming requests. The server only runs when the CLI arg is not passed.
• bubbletea: The primary library used to create the TUI. Charmbracelet's TUI toolkit.
• lipgloss: Charmbracelet library for stylish TUIs
• bubbles: The building blocks to the bubbletea library. bubbles is to bubbletea what a component is to a web application.
• godotenv: To access protected variables (like the OpenAI API Key) via a .env file
• go-openai: A library containing tooling for easy set-up and configuration for OpenAI clients with OpenAI-specific interfaces/types

Planned Improvements

• "Dockerizing" the application
• Better Log Implementation/Separation
• Compatibility with more LLMs (Claude is next)
• Enhanced Memory
• Retrieval-Augmented Generation Tooling
• Enable Performant Interfacing with ORM Databases (like Postgres) and/or Key-Val stores (like Redis)
• LangChain-like tooling
• Multi-stage and/or recursive/iterative prompting
• Potentially a Web Application to interface with the HTTP server
• Make it possible to host true web server with nginx implementation
• AND MORE!

Contact Me

Email: cam@fluenze.ai

Open to other suggestions! Please contact me if you are interested in contributing or open a PR!