HostedGPT offers an open-source, self-hostable alternative to ChatGPT, targeting users who want to run a polished AI chat interface locally or on their own servers. It provides a familiar user experience with support for multiple LLMs like GPT-4 and Claude 3, and features like conversation editing and custom assistants, all managed through a Ruby on Rails backend.

How It Works

This project is a Ruby on Rails application designed for ease of deployment and use. It leverages a PostgreSQL database for storing conversations and user data. The core functionality revolves around interacting with various Large Language Models (LLMs) via their respective APIs, allowing users to switch between models and custom assistants within a single conversation thread. The architecture prioritizes a user-friendly interface with mobile support and flexible authentication options.

Quick Start & Requirements

• Local Development: Requires Docker and Git. Run docker compose up --build in the project root.
• Deployment: Options include Render (free tier with limitations, paid upgrade for persistent database), Fly.io (approx. $2/month), Heroku ($7-$10/month), or self-hosting with PostgreSQL and Puma.
• Dependencies: Ruby, Node.js (for asset compilation), PostgreSQL, ImageMagick (for local non-Docker setup).
• API Keys: Requires OpenAI, Anthropic, or Groq API keys for LLM access.
• Documentation: HostedGPT Documentation

Highlighted Details

• Supports GPT-4, Claude 3, and Groq models within a single interface.
• Allows switching assistants mid-conversation and editing previous messages.
• Offers multiple authentication methods: Password, Google OAuth, Microsoft Graph OAuth, and HTTP header authentication.
• Includes experimental features like voice support and Google Tools integration (Gmail, Tasks).

Maintenance & Community

The project is led by an experienced Rails developer and actively seeks contributors. Development is ongoing, with recent updates (v0.6) including message editing, PWA support, and improved markdown rendering.

Licensing & Compatibility

The project is licensed under the MIT License, permitting commercial use and integration with closed-source applications.

Limitations & Caveats

The Render free tier database stops working after 90 days, requiring a paid upgrade. System tests do not currently function within Docker. Some experimental features like voice and Google Tools are noted as buggy or in development.