Now we have awesome models like R1 that rival o1 being 100x cheaper, gemini flash 8b with a 1m context length and support for basically every file type, then we have sonnet 3.5 that peaks at coding tasks!
Rather than using a framework that makes it easy to switch models and manage schemas, I built a tool that automatically combined the best of all models with a suite of powerful tools like AI scrapers, OCR, web search, etc
How it works? On each execution the tool will first auto-select the best model based on your prompt objective then it will tool call the relevant tools if required for more context or to perform a specific action. Then it will automatically decide to run all this through a reasoning model (R1) depending on the complexity of the prompt. Lastly, it will run all outputs combined on the auto-selected LLM and return the response.
View all models available which in my opinion are some of the best in the market right now: https://github.com/JigsawStack/omiai/blob/94ea56a9f953b4d229...
View the repo here: https://github.com/JigsawStack/omiai
Tech used: - Vercel's AI SDK as the underlying framework - JigsawStack for the AI tools and embedding model
Note: I built this in 24 hours, so It's still pretty early and expect breaking changes
The goal is to keep it opinionated meaning never giving the ability to pick models and limit the amount of configurations/tools. We'll only replace models with better ones and are unlikely to add. Your opinions matter too, so if you think something can be made even simpler or a config setting needs to be added, feel free to open a discussion/pr and we can take it from there!