Why every tool I build will be local-first

I spent last night writing a manifesto that boils down to one line: **Your prompts never leave your device.** Not a promise. An architectural guarantee. The app runs on your machine. API calls go directly from your browser to the AI provider. My server only touches the billing layer. I never see your content. Here's why this matters and why I think the current model is broken. ## The gatekeeping problem Want to try an AI tool? Create an account. Verify your email. Pick a plan. Enter your credit card. Now you're locked in — paying $15/month whether you use it once or a hundred times, getting promo emails you never asked for, with your prompts stored on servers you can't audit. That's not a product. That's a trap dressed up as a product. ## What I'm building instead **No accounts.** You open the tool and use it. No username, no password, no "sign up to continue." **No subscriptions.** Pay for what you use, when you use it. Prepaid credits. Top up when you want, stop when you want. Like a transit card — not a gym membership. **No data harvesting.** I don't store, log, or mine your inputs. My backend handles credit deduction. The content of what you're asking bypasses me entirely. ## The business model I'm a utility reseller. AI providers are the power generators — I buy wholesale tokens, sell retail. The margin between what I pay per token and what I charge is the business. Simple, transparent, scales with usage. Free trial on launch. Rate-limited, using my API key, so you can experience the tool with zero friction. When you want more, you buy credits through a one-time Stripe payment. No account created. No subscription initiated. ## Where this goes In five years, on-device inference will handle most everyday AI tasks. The cloud API becomes reserved for heavy lifting. When that happens, "your prompts never leave your device" becomes even more true — and my costs drop because I'm not paying for API calls. The account-per-tool model will feel as outdated as remembering your AOL login. Device-level identity and native payment rails will replace it. I want to be building for that future. Starting now. ## First up: SpecifyThat SpecifyThat is the first tool getting rebuilt with this philosophy. Today it's a server-side app where my API routes see every prompt. The rebuild flips that: API routes become thin proxies that forward to OpenAI without logging or storing what you send. My server never writes your prompt to a database or a log. Same tool, fundamentally different architecture. More on that soon.