Self-hosted ChatGPT for the API-key paranoid

What it does

This is a two-piece deployment: a Vue3 chat interface and a FastAPI backend that proxies all OpenAI API calls. The frontend never touches ChatGPT directly; the backend holds the API_KEY and streams responses back. It supports continuous conversation, gpt-3.5-turbo, and optional HTTPS_PROXY for regions where OpenAI’s endpoints need a tunnel.

The interesting bit

The author explicitly built the backend because frontends shouldn’t carry API keys. That’s the whole architectural premise — not a fancy model wrapper, not RAG, just “don’t leak credentials to the browser.” The frontend can still override the backend key if users want to bring their own, which is a pragmatic concession to multi-tenant or demo setups.

Key highlights

• Backend proxies all OpenAI requests; key stays server-side
• Frontend can inject its own API_KEY, overriding the server default
• Docker image published (miclon/chatgpt:latest) for one-container deploys
• Supports proxy via HTTPS_PROXY environment variable
• Continuous conversation and gpt-3.5-turbo model support

Caveats

• README is sparse on features beyond basic chat; no mention of streaming, rate limits, or auth beyond the API key
• The “balance” screenshot suggests some accounting UI, but the README doesn’t explain how it works or if it’s just parsing OpenAI’s usage endpoint
• No tests, no CI details, no license file mentioned

Verdict

Worth a look if you need a quick self-hosted ChatGPT UI and your threat model includes “API keys visible in browser devtools.” Skip it if you need multi-user auth, usage quotas, or anything beyond a single-key proxy.