A free-LLM API that runs on borrowed horsepower

What it does

Free-GPT4-WEB-API is a Flask-based HTTP server that wraps the g4f library (gpt4free) to expose free LLM providers—Bing, DuckDuckGo, and others—as a single local API. You send it a GET request with a ?text= parameter, it returns plain text. There’s also an optional browser-based settings panel for switching models, providers, and toggling history or proxies.

The interesting bit

The project knows its foundation is shaky and builds guardrails around that. It adds automatic provider fallback, health monitoring, a blacklist for dead providers, retry logic with 30-second timeouts, and even “virtual users” to distribute requests. It’s essentially reliability engineering applied to an inherently unreliable supply chain.

Key highlights

• Supports GPT-4, GPT-4o, DeepSeek-R1, Gemini, Claude via free provider proxies
• Docker image published with ~unspecified pull count (badge exists, no number visible)
• Optional web GUI for configuration; password-mandatory in Docker
• File upload support via multipart POST
• Siri shortcut integration for iOS voice queries
• GNU GPL v3.0 licensed

Caveats

• Depends entirely on g4f/gpt4free, which scrapes or reverse-engineers free tiers; provider breakage is expected and documented
• “Unlimited requests” and “no rate limiting” are aspirational claims about the wrapper, not the underlying providers, which absolutely rate-limit or block
• Security notes in the README suggest prior authentication bypass issues; current version logs attempts and hardens against bypasses
• Demo server “can be overloaded and may not always respond”

Verdict

Useful if you need a quick, self-hosted LLM proxy for low-stakes prototyping, personal automation, or Siri integration—and you accept that answers may arrive slowly, wrongly, or not at all. Skip it if you need reliability, data privacy guarantees, or anything production-facing.