What you're probably wondering.

No. Five agents argue per JD — buyer psychology, archetype routing, specifics-density, fingerprint scrubber, fact guard. Built on Sonnet 4.5 with Gemini Flash for cheap stages, running in production since May 2026. The argue + lock + scorer loop is the actual product, not the LLM call.

Hard-coded fingerprint scrubber catches 30+ LLM-tells — corporate fluff (leverage, robust, seamless), soft-adverb openers (Importantly, Notably), tricolons (X, Y, and Z), semicolons, closer pleasantries (looking forward, happy to discuss). Em-dash cadence preserved — Talha actually uses em-dashes 4-8 times per letter. Specifics density must be ≥9 per 100 words. Auto-refines once. Blocks if it still fails.

Fact guard regex-blocks every fake metric, star rating, percentage claim, fake testimonial quote, unverified named client, and invented revenue figure before the LLM ever finishes. Then a Flash cross-check compares against the verified-facts whitelist. If a claim isn't in there, the draft can't make it through.

Studio rate, not freelancer rate. One Upwork deal you'd otherwise miss pays for several months. The math works for operators earning $50K-$200K/year who lose deals to slow replies and template-looking proposals. If you're earning $15/hr, this isn't your tool.

/work has all seven live production builds — Dr. Bryant Medical (patient portal, Twilio OTP, Stripe, D1 13 tables), RentCore (rental ops platform), RepuShield (GBP reputation SaaS), SignalForge (multi-brain trading intelligence), Closeline, themeknock.net, ThemeKnock Studio. Click any one and use it right now. On a 25-minute call I'll screen-share the codebases too.

No. Closeline drafts; you review and submit. It does not log into Upwork, scrape their site, message clients, or interact with the platform in any automated way. You stay in the loop on every reply.

Closeline routes through OpenRouter with model fallback. If Sonnet is down, it can degrade to Gemini Flash for the cheap stages. The cover letter scorer + auto-refine path requires Sonnet — so during a hard outage, drafts queue rather than ship low-quality.

Yes. Every proposal exports as a Claude Code-ready bundle — the JD, the council output, the cover letter, the HTML proposal page, the reply history, and the typed memory. Your data, your repo, day one. Cancel anytime, you keep all of it.

First proposal is shipped in about 4 minutes from JD paste to ready-to-send. Reply rate improvement is faster to measure than close rate — most operators see lift on first cohort of proposals within a week. Close rate improvement compounds with typed memory over 30-60 days.

Closeline runs itself — you don't need to write code. The 25-minute onboarding call walks you through pasting a JD and reviewing the output. If you're stuck, I personally review your first 5 drafts (trial month includes this).