singularis

Reading https://ai-2027.com/ after the whole Mythos/Fable 5 saga hits completely different.

Check out SearchAugmentedLLM, a new project that empowers Large Language Models by integrating real-time web search capabilities. This tool performs Google searches based on user queries, extracts relevant content, and ranks it for contextualized responses. Ideal for Retrieval Augmented Generation (RAG) applications, it's currently in beta and comes with REST API support. Great for developers looking to improve LLM accuracy with up-to-date information! Explore more on GitHub: https://github.com/EliasPereirah/SearchAugmentedLLM

It's hard to boycott Google, but if I were an employee there I would quit. https://www.youtube.com/watch?v=7aykDynnZpc

The company raced to beat Amazon to sell AI tools to the Israel Defense Forces.

This is a open source project allows you to use various models, including those from OpenAI itself and some free ones like Google Gemini, Cohere, SambaNova, Groq, and Cerebras. It's worth noting that Google Gemini is among the best in the LLM Arena.

Why is it free? Google provides a free API with up to 1500 requests per day if it's a Flash model, and 50 requests for the Pro model.

Google isn't the only one; there are several other free options.

Interface: https://eliaspereirah.github.io/OrionChat/

GitHub: https://github.com/EliasPereirah/OrionChat

All have good models, but i agree with jtbx, Gemini it is very good choice. Sometimes i use gemini-2.0-flash-exp because the gemini-exp-1206 do not work all the times. I think Google is about to release new models

Example: SELECT x,y,z WHERE MATCH(content) AGAINST(:query) AND author = 'jhon'

You can use several other models, including local models using models available in Ollama.

Thanks! I'm glad you're enjoying it. You can add custom prompts directly in the interface without editing prompts.js.

I added a robots against facebookexternalhit/1.1 and after a while the hits stopped. An interesting point is that my website had an error regarding the canonical url and the og:url meta, which basically created a link to a non-existent page. If the page /x was accessed, og:url would be /x/y and if /x/y was accessed then og:url would be: /x/y/y and so on, perhaps this was one of the causes. Although Facebook accesses were repeated on the same pages already accessed.