Build AI Agents That Find Clients and Build Their Websites
Venari turns an idea into a website. You describe it, scouts search Pinterest and real websites in your field, you love the designs you want or bring your own pictures, answer a few questions, and your own Hermes writes a creative brief and builds the website. Rebuild a design you upload, change a version with a sentence, find local businesses that need a better website and send them the one you made for them.
42 prompts · 6 parts · copy-paste ready
// Contents
Part 01 / 06 — The Server and the Studio
Part 02 / 06 — The Design Search
Part 03 / 06 — The Builder
Part 04 / 06 — Your Websites
Part 05 / 06 — Clients
Part 06 / 06 — Finishing Touches
Prerequisites
- ▸A Linux server that stays on, with Hermes Agent installed and signed in to a model (hermes -z "say ok" answers)
- ▸ffmpeg and Node.js 18+ (sudo apt install ffmpeg nodejs npm)
- ▸The Venari studio template, brought in at prompt 2
- ▸Before Venari scores a business that Google Maps lists without a website, it now checks whether the business has one anyway. If it finds the site, the lead is marked "Has a website Google Maps does not list" and scored on that site, so you no longer pitch a new website to someone who already owns one.
The Server and the Studio
The Contract
Claude Code
What to expect: a short plan and an empty server.py that starts and answers on port 8886.
We are building Venari, a design search studio and website builder, in ~/venari. Rules
for every step from now on:
- One Python 3 server, server.py, standard library only (ThreadingHTTPServer). No
framework, no pip packages. Helpers go in modules next to it.
- It binds 127.0.0.1:8886 and never anything public.
- Data lives in ~/venari/data: runs/, img/, sites/, clients/, uploads/, settings.json,
keys.json (mode 600). Writes are atomic (temp file, then rename).
- Every /api/ answer is JSON. A problem is {"error": "one plain sentence"} with status
400; never a crash, never a dropped connection.
- The studio page is a finished download that nobody edits. When I say "the template
expects", read how the page calls that endpoint and match it exactly.
- Long work (searches, briefs, builds) runs in a background thread and is saved as it
goes, so a restart never loses it and the page can poll.
Make the empty server, start it, and show me it answers.Bring In the Studio
Claude Code, with the template saved as `~/venari/studio.html`
What to expect: the studio opens at http://localhost:8886 (empty, with no errors).
Show full prompt
I saved the studio template as ~/venari/studio.html. Serve it at / and /index.html. Also
serve:
- /live-version: {"v": a hash of studio.html}, so an open page reloads itself after an
update;
- when serving /, add to the head a manifest link, an apple-touch-icon link and the
apple-mobile-web-app-capable meta tags, and serve /icon-192.png and /icon-512.png;
- /img/<file> from data/img and /uploads/<file> from data/uploads (pictures only);
- a web app manifest at /manifest.webmanifest named Venari, so phones can add it to the
home screen.
Then open the page, read its script and list every /api/ endpoint it calls, grouped by
screen. Keep that list in API.md: it is our checklist for the rest of the build.Talk to Your Hermes
Claude Code
What to expect: python3 -c "import ai; print(ai.ask('Reply with: ok'))" prints ok.
Show full prompt
Write ai.py, the only way Venari talks to a model. Everything goes through my Hermes, so
any model it can run works:
- ask(text, images=(), model=None) runs `hermes chat -Q -t todo -q <text> -m <model>
--provider <provider>` with the chosen model (see prompt 4), drops any line starting
with "session_id:" and returns the reply. One picture goes in with --image; several
are joined into one numbered sheet with ffmpeg first. Text too long for a command line
goes in a temp file, and Hermes is told to read it with its file tool.
- ask_json(text, images=()) asks for JSON only, finds the JSON in the reply, and if it
does not parse, asks once more with the parse error. Raise a plain error after that.
- A timeout of 10 minutes, and a clear error if Hermes is missing or does not answer.
Test both with a real call, one with a picture.All 42 prompts are . Or skip the setup:
Complete Kit (Prompts, Template and Auto-installer)
Dashboard installed for you.
- The prompts and template for this dashboard
- The one-command auto-installer
- Install on up to 5 machines
- No subscription
One tutorial only. Pay once, keep it.
Membership
This installer now, the whole library from month 4.
then $10/month, cancel anytime
- This installer plus 2 more paid tutorials you pick
- Every new tutorial, the day it drops
- The whole library from month 4
- 15 dashboard plugins
- Private Telegram group with other builders
275+ members
Prompts + Template
Build it yourself, step by step.
- All prompts from the video
- The finished dashboard itself. The prompts build the engine behind it, the template is what you see and use.
- Follow along at your own pace
No auto-installer included.
Your first 2 months of membership cost less than one Kit, and include this installer plus 2 more you pick.
Get all 42 prompts by email
No card needed.
Want them as a single Markdown file plus the dashboard template (.html)? We'll email you a 6 digit code to confirm.
// What you're building toward
Your own design studio: an idea goes in, real designs come back, and your Hermes builds the website and helps you find the client who needs it.
Stuck on a step? Drop a comment on the video, or book a session and we'll walk through it together.
Usage & License
Use the prompts and downloadable templates on this page in your own personal and commercial projects. What's not allowed: reselling, repackaging, redistributing, or republishing this material — on YouTube, Gumroad, paid courses, newsletters, or anywhere else. Send people to komputermechanic.com instead. Violations get reported and taken down.