LocalLens: private document Q&A that runs 100% on your machine

LocalLens is a new private document Q&A tool that runs entirely on your machine, with no API costs or subscriptions. It was created to address the problem of confidential documents that cannot be uploaded to services like ChatGPT due to non-disclosure clauses. Freelancers, HR and accounting firms, and clinic staff are potential users who need to review sensitive files without risking data breaches.

The tool is a command-line application where users can ask plain English questions about a document and receive answers in seconds. The AI model, retrieval system, and inference engine all run locally on the user's computer, ensuring complete privacy. The demo shows how to use the tool with a 34-page client contract and asks about the notice period for terminating the agreement. The answer is provided in real-time, citing the relevant sentences from the document.

The developer emphasizes that the tool cites its sources, allowing users to verify the answers before acting on them. If the answer is not found in the document, the tool will clearly state that it was not found, rather than generating a false response. This feature sets the tool apart from others that may hallucinate answers and potentially create liability.

The business model proposed is a one-time purchase for a polished desktop app with support for PDF and DOCX files, and multi-document search. The open-source core is available under the MIT license, with paid packaging for convenience. The tool relies on open weights models, such as Qwen2.5-0.5B-Instruct, ensuring transparency and auditability. The current version focuses on keyword-based retrieval and is limited to .txt files, with future plans for PDF/DOCX parsing and support for larger models.