yagyeshVyas / personalforge

What is personalforge?

PersonalForge lets you build your own offline AI from documents like PDFs, Word files, Excel sheets, or code—without writing code. Upload files via a local Python Flask web app, pick a thinking mode (coder, deep thinker, or factual), generate Q&A training pairs, then use the auto-created Colab notebook to LoRA fine-tune a model on free T4 GPU and export a GGUF file. Run it forever offline in LM Studio or Ollama, with baked-in knowledge plus optional RAG for big datasets.

Why is it gaining traction?

It stands out by delivering a fully private, zero-cost pipeline to build your own LLM or RAG setup—no cloud leaks or subscriptions like commercial RAG tools. Developers dig the three specialized modes that shape responses (e.g., code examples with best practices), nine Unsloth-optimized models from 1B to 7B params, and one-click Colab export that handles rsLoRA, quantization, and early stopping. The hook: turn personal notes into a custom agent faster than manual fine-tuning.

Who should use this?

Solo devs building their own AI agent akin to a GitHub Copilot for private codebases or docs. Researchers or students with textbooks/papers wanting an offline tutor. Professionals like lawyers or doctors needing a factual reference from confidential files, without data leaving their machine. Anyone tired of generic chatbots who want to build own AI from their exact knowledge base.

Verdict

Try it for personal projects if you have Colab access—solid docs and intuitive flow make it beginner-friendly despite 46 stars and 1.0% credibility score signaling early maturity. Skip for production until more benchmarks and voice/multi-GGUF features land on the roadmap.

(198 words)