An open source AI search engine. Powered by Together.ai.
If you want to learn how to build this, check out the tutorial!
- Next.js app router with Tailwind
- Together AI for LLM inference
- Llama 3.1 8B and 70B for the LLMs
- Bing / Serper API for the search API
- Helicone for observability
- Plausible for website analytics
- Take in a user's question
- Make a request to the bing search API to look up the top 6 results and show them
- Scrape text from the 6 links bing sent back and store it as context
- Make a request to Llama 3.1 70B with the user's question + context & stream it back to the user
- Make another request to Llama 3.1 8B to come up with 3 related questions the user can follow up with
- Fork or clone the repo
- Create an account at Together AI for the LLM
- Create an account at SERP API or with Azure (Bing Search API)
- Create an account at Helicone for observability
- Create a
.env(use the.example.envfor reference) and replace the API keys - Run
npm installandnpm run devto install dependencies and run locally
- Move back to the Together SDK + simpler streaming
- Add a tokenizer to smartly count number of tokens for each source and ensure we're not going over
- Add a regenerate option for a user to re-generate
- Make sure the answer correctly cites all the sources in the text & number the citations in the UI
- Add sharability to allow folks to share answers
- Automatically scroll when an answer is happening, especially for mobile
- Fix hard refresh in the header and footer by migrating answers to a new page
- Add upstash redis for caching results & rate limiting users
- Add in more advanced RAG techniques like keyword search & question rephrasing
- Add authentication with Clerk if it gets popular along with postgres/prisma to save user sessions
- Perplexity
- You.com
- Lepton search