This isn't a pitch deck claiming an empire. It's an honest readout of what one operator and a fleet of AI build-agents have actually shipped — measured, not estimated. The page renders itself from a live collector, so the numbers stay true. It's the proof, not the promise.
Measured Estimate Indicative
Pulled at render time from story-assets/tech-metrics.json. Each tile shows its source. Tokens and build-minutes are real measured per-subagent telemetry; dollar costs are tagged estimates.
The per-agent figure above is the marginal cost of one build run — not what the whole company cost to build. Here is the real, measured spend — the 2025–26 software/AI window and the ~$51K all-time studio build — from the filed LLC P&L and the founder's card statements. Tagged, reconciled, and honest about what is still year-to-date.
Computer & Internet $7,065 + Dues & Subscriptions $6,778. Total opex $15,272; net loss −$15,820. Measured · accounting
Anthropic $5,025 · Manus $4,802 · ElevenLabs $2,363 · infra/tools the rest. Through Jun 12. Measured · Amex
2025 booked + 2026 card YTD — the measured software/AI window. A bootstrapped build, not a funded burn. Measured
$32.5K AppSumo lifetime deals (since 2023) + ~$18.5K measured API/AI/infra. Measured
~2,700 hrs at $63.59/hr (founder W-2 opportunity cost) ≈ $172K of sweat equity on top of cash. Time = estimate
Honesty: 2025 is accounting-grade from the filed Vital Webmaster LLC P&L; 2026 is card-measured and year-to-date, so it will grow. Founder-time is an opportunity-cost estimate (real W-2 rate), not a cash outlay. The widely-feared "AI bill" is small and measured: Anthropic ≈ $5K in 2026, not the five-figure guess. Detailed entity financials are being formalized in CloviFinance.
This grid is generated by scanning real package.json deps, Python imports and live CLI probes (PM2, FFmpeg) — not a wish-list. AI model names are listed as our own tools.
Own code only, across named repo roots, excluding node_modules / dist / vendored deps. Counted live — vendored libraries are not claimed as ours.
Charts on this page are inline SVG/CSS for now. TODO: swap the language-mix bar and stat sparklines to live CloviCharts embeds once the static-embed endpoint ships — dogfooding our own viz tool is the point.
Not just code — measured, embedded knowledge bases the agents read from and write back to. Counts are live entry counts, not round numbers.
A simple, honest architecture: an orchestrating operator dispatches specialised build-agents; agents read shared RAG/KB layers and emit products, pages and data — which feed back into the knowledge layer. This page was produced by that exact loop.
Services health: — · — services OK monitored
A concrete velocity proof — not a marketing claim. These are the agents logged in a single recent work window (including the CloviSlider research wave), measured straight from the build ledger.
CloviTek's accessibility scanner, performance harvester and optimization KB run against clovitek.com itself. The honesty rule applies here too: before/after is shown only where we have the receipts.
Heavier render, unminded Core Web Vitals — the baseline our own harvester flagged.
66 mechanical perf/a11y/SEO fixes in the KB, applied where verified-safe.
Detailed CWV before/after reports live alongside the homepage QC artifacts. TODO: link the specific homepage-qc CWV report once its public URL is finalized — we don't publish a number here we can't point at.
Every figure on this page is emitted by a single re-runnable collector — /root/scripts/gen_tech_metrics.py — which reads real sources: the agent build ledger (measured per-subagent token + duration telemetry), the PM2 process registry, a live HTTP probe of public product URLs, a cloc/find+wc sweep of our own repos, and live entry counts of the RAG/KB stores. It writes story-assets/tech-metrics.json, which this page fetches at runtime — so re-running the collector updates the page, and nothing is hand-typed.
Provenance tags: Measured = real telemetry or a live count. Estimate = derived (e.g. dollar cost from tokens × blended rate; tokens themselves are measured). Indicative = LOC, which counts only our code over named roots. Anything the collector can't measure honestly is listed in the JSON's flags array rather than invented. Same standard as The Numbers.
Re-run any time: python3 /root/scripts/gen_tech_metrics.py
