ZeroLLM launches TinyLlama-based RAG assistant

The "from scratch" framing is doing some heavy lifting here: the [Reddit announcement](https://www.reddit.com/r/LocalLLaMA/comments/1s6tn4q/built_an_open_source_llm_from_scratch_zerollm/), [GitHub README](https://github.com/ashwin123-git/ZeroLLM/blob/main/README.md), and [demo site](https://zerollm-ai.vercel.app/) all describe ZeroLLM as a fine-tuned TinyLlama 1.1B RAG assistant, so the interesting part is the product wrapper, not a new foundation model.

• –The training mix, OpenHermes 2.5, Dolphin Coder, and Orca Math, points at a practical assistant tuned for chat, coding, and reasoning.
• –Real-time web search, code generation, file uploads, and chat history is a sensible feature set for a lightweight assistant that wants to feel current without needing a giant model.
• –GPL-3.0 keeps the project easy to inspect and fork, which is probably the main appeal for builders.
• –The small base model keeps it approachable, but it also means retrieval quality and UX will matter more than raw model power.