Show HN: TinySearch - Give small LLMs fast web access without context bloat (github.com)

TinySearch, a newly announced local-first web research engine, aims to enhance the capabilities of small language models (LLMs) by enabling them to perform fast, efficient web searches without the hassle of managing extensive infrastructure. Unlike traditional search engines, TinySearch emphasizes a lightweight approach where it directly searches the web, reranks results, crawls selected pages, and extracts relevant data to construct a source-grounded prompt for LLMs. This process eliminates the need for full webpage context, which can lead to bloated responses, making it ideal for local agents, personal workflows, and prototyping. This innovation is particularly significant for the AI/ML community as it prioritizes efficiency and relevancy in web-based research, tailored for smaller systems. TinySearch supports ONNX embeddings and OpenAI-compatible APIs, allowing for various embedding models to be utilized effectively. By returning concise prompts with attached source URLs, it enhances the accuracy and reliability of the LLM’s responses. Additionally, its minimalistic design—void of hosted dashboards or analytics—ensures a streamlined user experience for those in need of focused research capabilities without the overhead of a full search backend.