LibreChat for use by attorneys discussed at length here

[sethchandler]

Confidentiality solved contains an 8500 word discussion of why attorneys concerned about confidentiality in using large language models should strongly consider LibreChat in conjunction with OpenRouter. Contains an extensive "how to" on installation issues intended for people (like most attorneys) who are not technically inclined.

TL;DR: The post outlines a practical, lawyer-friendly privacy architecture for using frontier LLMs without leaking client confidences. Run LibreChat locally (localhost:3080) as the UI/router, pipe all requests through OpenRouter configured to allow Zero Data Retention (ZDR) endpoints only, and never touch a vendor web app. With ZDR on both hops, providers see traffic from OpenRouter—not you—and nothing is stored server-side; chats can also be “temporary” to avoid local persistence. The guide walks through setup (Docker vs. git install), obtaining and funding an OpenRouter API key, enforcing ZDR in settings, enabling web search/file uploads, and common LibreChat features (agents, artifacts/canvas). The result is model-agnostic (GPT-5, Claude 4.5, Gemini 2.5, etc.), inexpensive (pennies per query), and plausibly consistent with attorney-client privilege because data never leaves your machine except via a ZDR conduit. Caveats: requires moderate technical competence, some cost, and your own ethical due diligence; the author explicitly disclaims legal advice.