Defluff

Strip AI-generated padding from emails. Your keys, your models, no servers.

Corporate email is being drowned in AI-generated filler. Half of every message is restated context, "I hope this finds you well," and softening qualifiers that exist only because a model was asked to pad a one-line request into four paragraphs. Defluff adds a one-click "De-Fluff" button next to emails in Gmail, Outlook Web, and LinkedIn messages. Click it — the message collapses into 3–5 bullets of what the sender actually wanted, plus a verdict (ACTIONABLE / RESPONSE-NEEDED / FYI / NOISE) and a reversal of the likely prompt behind the email. What makes Defluff different is the architecture. ZERO RETENTION BY DESIGN There is no Defluff backend. Your email body and your API key never leave your browser for any infrastructure we run. Every De-Fluff click goes straight from your browser to the LLM provider you picked. BRING YOUR OWN KEY Use Anthropic (Claude), OpenAI (GPT), Google Gemini, or a local Ollama server. Your choice, your API key, your bill. Switch at any time. HOST-SCOPED PERMISSIONS Defluff only injects into mail.google.com, outlook.office.com (and Microsoft's other Outlook origins), and — only if you opt in via the options page — linkedin.com. Never "all URLs." You can read every host permission in the manifest before you install. OPEN SOURCE (MIT) Around 2,000 lines of TypeScript total. Auditable in an afternoon: github.com/FinAegis/defluff NO TELEMETRY No analytics SDKs, no crash reports that ship request bodies, no usage tracking. The extension makes exactly one external request per click — to the LLM provider you configured. TYPICAL USE • A 400-word "circling back" email → two bullets + NOISE verdict. • A status update buried in three paragraphs → three bullets of what's on fire. • A sales pitch disguised as a personalized intro → the reversed prompt makes it obvious. COST $0 for the extension. Your provider bills you per request directly — typically under $1/month on Claude Haiku or GPT-4o mini, even for heavy users. $0 total if you point Defluff at a local Ollama model. WHO IT'S FOR • Anyone who reads more email than they write. • Teams whose CISO has already banned Grammarly, Otter, Shortwave, and Superhuman on data-residency grounds — Defluff has nothing to ban, because no data flows through a Defluff server. • Self-hosters who want AI assistance without handing content to a SaaS. PRIVACY POLICY github.com/FinAegis/defluff/blob/main/PRIVACY.md Short version: no backend, no analytics, no account. Your email content goes only to the LLM provider you configured, using the API key you control. SUPPORT • Issues & feedback: github.com/FinAegis/defluff/issues • Source: github.com/FinAegis/defluff (MIT)