Self-hosted chatbot console with a two-step promise

What it does

OpenChat is a self-hosted web console for managing multiple customized GPT chatbots. It grounds conversations in external context—PDFs, websites, or an entire GitHub repository—using a vector database like Qdrant or Pinecone. The backend has been rewritten in Python Django, while the frontend is currently mid-migration to Next.js and TypeScript.

The interesting bit

The project pitches itself as an “everyday user” tool with a two-step setup, but the README quickly devolves into Docker Compose, environment variables, and API-key juggling. The real value is in the document ingestion: it can chew through a 400-page PDF or an entire GitHub repository and let a bot answer questions against that memory, which is the tedious part most teams don’t want to build from scratch.

Key highlights

• Supports GPT-3/4 via OpenAI or Azure OpenAI, with Llama2 support recently added
• Vector-store memory using Qdrant (preferred) or Pinecone for large-context retrieval
• Embeddable JavaScript widgets for websites and internal tools
• “Pair programming” mode ingests a full codebase via the GitHub API for repository Q&A
• Prebuilt Docker images available for slower connections

Caveats

• The “two-step setup” is marketing, not reality: you need Docker, API keys, and careful env-file configuration
• The frontend is actively being rewritten in Next.js/TypeScript, so the current UI is transitional
• True offline operation and open-source LLM drivers remain on the roadmap, not in production

Verdict

A sensible choice if you want a self-hosted, multi-bot interface over documents and code, and don’t mind managing infrastructure. If you’re looking for a managed, click-and-go SaaS, this is not it.