// the find
IntelligenzaArtificiale/Free-Auto-GPT
Free Auto GPT with NO paids API is a repository that offers a simple version of Auto GPT, an autonomous AI agent capable of performing tasks independently. Unlike other versions, our implementation does not rely on any paid OpenAI API, making it accessible to anyone.
A langchain LLM wrapper that gets ChatGPT/Bing/Bard/HuggingChat responses by scraping web session cookies instead of calling paid APIs, used to run toy AutoGPT/BabyAGI/CAMEL agent loops for free. It's aimed at people who want to poke at autonomous-agent patterns without an OpenAI key, not at anyone building something that needs to stay working.
The LLM wrapper abstraction is genuinely clean - all four backends expose the same `llm(prompt)` call so you can drop any of them into a standard langchain agent/tool chain with no other code changes. The example agents (AutoGPT, BabyAGI, CAMEL, a 60-line custom agent with web search + Python REPL + Wikipedia) are short and readable as reference implementations of those patterns. There's at least a crude rate-limit counter on the ChatGPT wrapper acknowledging the 50 req/hour ceiling it's working around.
The whole project is built on scraping session cookies and reverse-engineering private endpoints for ChatGPT, Bing, and Bard - this is against those services' ToS and breaks every time the vendor changes their frontend, which is most of why reverse-engineered wrappers like this have a short shelf life. No commits since June 2024: Bard has since been folded into Gemini and Bing Chat's auth flow has changed, so a meaningful chunk of the four backends is likely dead on arrival today. There are zero tests anywhere in the repo. The same FreeLLM wrapper code is copy-pasted three times (root, OtherAgent/, hfAgent/) instead of factored into one shared module, so any fix has to be applied three times.