Four paths into envvoy. Pick whichever matches what you already have:

1. You're already inside an AI assistant. Paste read envvoy.ai/skill.md and register my agent into Claude, Cursor, ChatGPT, or any AI tool that can fetch a URL. The assistant reads envvoy's spec, drafts your registration, and asks for the details it doesn't know. Full walkthrough in From your AI assistant below.
2. You want to describe your agent in plain English. Click “describe instead” on /onboard/start and type a one-line description. envvoy walks you through the interview from scratch, no URL, no repo, no technical setup required. This is the path most no-code and Replit/Lovable builders take.
3. You have a GitHub repo. Paste the URL on /onboard/start. envvoy reads the repo's name + description, asks you four quick questions to fill what we couldn't pull, and drafts your site.
4. You have a hosted URL (landing page, marketplace listing, anything with an og:image+ description). Same path. envvoy scrapes what's on the page, confirms with you, and proceeds.

Either way you'll land on envvoy.ai/your-name with a full marketing page in under five minutes.

direct API access: live. POST to api.envvoy.ai/v1/register with the field schema at /skill.md. For the conversational path see path 1 above and From your AI assistant below.

If you're already inside Claude, Cursor, ChatGPT, or another AI assistant, envvoy can register your agent from that conversation. No forms, no dashboard. You paste one prompt, the assistant reads envvoy's spec, and drafts the registration for you.

Step 1. Paste this into your assistant

That's the trigger. The assistant fetches envvoy.ai/skill.md, which is the full spec: field schema, palette IDs, font pair options, slug rules, validation errors. The whole registration surface in one file.

Step 2. Answer a few questions

The assistant asks about your agent (name, tagline, what it does, who it's for, a couple of use cases). Give it as much detail as you can. Every additional signal improves the generated brand kit and how AI engines cite you. If you don't know an answer, just say so and the assistant will pick a reasonable default from skill.md.

Step 3. Verify your email once

The assistant POSTs to https://api.envvoy.ai/v1/register. You get a 6-digit code by email. Enter it at envvoy.ai/verify or click the link in the email. Your site goes live at {your-slug}.envvoy.ai within a minute.

Which assistants work

Anything that can fetch a URL and call an HTTP endpoint. Confirmed with Claude (web, Desktop, Code), ChatGPT with browsing enabled, Cursor, and any MCP client with web fetch. If the assistant refuses to fetch skill.md, that's a permissions setting on your end, not an envvoy issue. Give the assistant explicit access to envvoy.ai and retry.

Integrating envvoy programmatically

For AI engineers wiring envvoy into a product rather than a conversational assistant, the full spec including field types, slug transformation rules, error codes, and rate limits lives at envvoy.ai/skill.md. POST directly to https://api.envvoy.ai/v1/register. One registration per email, 5 requests per IP per hour, 72-hour unverified-registration TTL.

Six required answers drive every piece of generated copy on your site, agent function, name, audience, benefit, color preferences, and logo plan. The interview surfaces them in the order that's easiest to think through, not the order they appear on the page.

Two behaviors that aren't obvious from looking at the screen:

• Generic-noun flag.If your audience answer is “developers” / “users” / “people,” we'll suggest you sharpen. Answer engines reward specific noun phrases.
• Optional second pass.After the six required answers, an optional gate offers two more questions, a second benefit + a real-world scenario. Skip or fill inline. They sharpen the copy but aren't required to launch.

After the interview you'll land on /onboard/brand-kit-pick. Three decisions:

Tagline

A one-line summary that sits under your agent's name on the public site. Type your own or leave it blank, we'll draft one from your interview answers. Either way you can edit it later.

Colors

Four colors: primary, secondary, accent, neutral. The palette is drafted from your “color preferences” answer by Claude. You'll see them in the live preview card at the top of the page. Fine-tune any swatch from the editor after launch.

Fonts

Three pair options: Editorial (Instrument Serif + Inter, italic, magazine), Technical (Inter Tight + JetBrains Mono, dense, engineering), Modern (Inter + Inter, clean default). Each preview card renders your agent's name in the actual pair so you see the choice, not just the label.

Live preview at the top of the page updates as you tweak anything. What you see is what ships.

If you picked “design one for me” in the interview, after signup you'll land on /onboard/pre-launch/logo. envvoy passes your brief + visual cohort match + style/personality picks to gpt-image-1 and returns a single icon mark.

You get three free generations per session. If the first attempt isn't right, two regen buttons let you steer the next attempt:

• another like this, mild variation of the current direction
• completely different style, aggressive push into a different visual register

Each generation takes ~15-20 seconds. The model has no random seed, so diversity comes from the prompt itself, click “completely different style” if you're getting variations on the same form.

What makes a good logo brief

envvoy's prompt enforces strong design principles automatically: zero text, exactly two colors, readable at 32px, abstract over literal, anchored to a visual cohort distilled from real AI-company marks. Your job is the input, specific descriptors of what your agent does and who it's for produce better marks than vague descriptions. If you wrote “an AI tool for stuff” the model has nothing to work with.

You can refine the logo from the editor after launch. The mark is yours, download as SVG/PNG from the brand-kit panel.

copyright note: the mark functions as your trademark through use in commerce. copyright treatment of AI-generated images is evolving, the US Copyright Office has indicated that purely machine-generated images may not be independently copyrightable. we don't make copyright claims about generated logos; consult counsel before relying on copyright protection for a generated mark.

After your site is live you can edit any field at envvoy.ai/edit/your-slug. The page auto-saves as you type, no save button to hunt for. The header status indicator switches between “saving…” and “✓ saved” so you always know where you stand.

What's editable:

• Agent name, click and type. Underlined + has a ✎ icon so you know it's a field, not a title.
• Tagline, same.
• Palette, four swatches with color pickers. Live preview on the right updates as you tweak.
• Fonts, same three pairs as the onboarding picker. Switch anytime.
• Hero / audience / differentiator / about paragraphs, direct edit on the body copy. Each has a “reset to AI version” button that restores Claude's polished draft if you rewrote it and changed your mind.
• Logo, “design a logo” button takes you back to the logo step with your current answers pre-loaded. Or upload your own.
• Visitor action (CTA). Try it / View on GitHub / Get notified / Coming soon / custom. The picker handles the destination URL conditionally so you can't pick “Try it” without setting where to.
• Deeper questions, about-section, real-world scenarios, jargon to use/avoid, who built this, why. None required; all improve the AI copy when filled.

Changes are saved as drafts. When you're ready, hit “publish changes” at the top or bottom of the page, the public URL updates within a few seconds.

AEO is answer-engine optimization. SEO got your site ranked in Google search results; AEO gets your agent named when someone asks ChatGPT, Claude, or Perplexity for a recommendation. They're different problems and they have different audiences.

envvoy's site template is structured to give answer engines what they need to confidently name your agent: a clear single-sentence description, a specific audience phrase, a concrete differentiator, and consistent metadata (Open Graph tags, JSON-LD, semantic HTML).

What we do for you automatically:

• Clean semantic HTML, proper <h1>, sectioned content, accessible labels.
• Open Graph + Twitter Card metadata on every page so when someone shares your link, the preview looks right.
• JSON-LD structured data describing your agent, the machine-readable layer that AI crawlers use.
• Plain-language copy. We strip marketing-speak (“revolutionary AI-powered platform”) from generated text because answer engines penalize it, not reward it.
• Concrete specifics over abstractions. “Scans 200 chains” beats “industry-leading coverage.” The Claude polish step is tuned for this.

Off-site work is tier-gated. The $50 ad-hoc bulk paid-directory submissionpushes you onto 25+ directories and rating sites in one go. Monitor & Improve ($19/mo) keeps submitting you to new directories, registries, and rating services as they appear. Backlink outreach + content amplification lives on Grow ($49/mo).

Your public URL is envvoy.ai/your-slug. Share that link anywhere. Twitter, Hacker News, Reddit, email, marketplace listings. The preview card (Open Graph) renders automatically with your logo + tagline.

The site is fully public; no auth wall. Visitors don't need an envvoy account to view your page. The only authenticated surface is your editor at /edit/your-slug.

Custom domain

Custom domain hosting (pointing your own myagent.com at your envvoy site) is included on Grow ($49/mo). On any other tier, a one-off $25 setup wires up your domain and ships a ready-to-merge GitHub PR. The free tier defaults to your-agent.envvoy.ai.

I pasted a GitHub URL but it didn't fill anything in

GitHub's public API caps at 60 requests per hour for unauthenticated traffic. If a lot of other envvoy customers are testing at the same time, we can hit that ceiling. If it happens to you, you can skip forward, the questions still let you fill those fields manually. The fix is on our side (we're authenticating).

The logo I generated doesn't look right

You have three free regens. Use them. Click “another like this” for variation on the current direction, or “completely different style” to push the model into a contrasting register. If none of three feel right, refine your interview answers (more specific = better logo) and try again, the answers feed the prompt.

The site copy reads bland

Claude can only work with what you gave it. If your interview answers were thin or generic (“helps people with stuff”), the output will be too. The fix: open the editor, click into the paragraph that's weak, write more specifically. Then click “reset to AI version” on the others, the next polish will use your richer input.

I've been live 6 weeks and no AI engine is citing me

This is the failure mode customers actually hit, and it's usually not a failure, it's normal early. What's normal:

• Weeks 1–2: answer engines have to crawl, index, and form an opinion of your page. Almost nobody is cited in the first two weeks unless they were already known.
• Weeks 3–6:first citations start showing up. They tend to come from specific long-tail queries (“python refactoring tool for indie devs”) before broad ones (“best AI coding tool”).
• Weeks 6+:if you're still at zero citations across ChatGPT / Claude / Perplexity, that's a real signal worth investigating. The most common cause: the site's copy is too generic. Open the editor and rewrite the hero, audience, and differentiator paragraphs with specifics, concrete names, real numbers, the exact problem you solve. The Claude polish step works downstream of your input; if the input is vague the output is too.

Off-site amplification is tier-gated. The $50 ad-hoc bulk-directory submission is the initial push to 25+ directories. Monitor & Improve ($19/mo) keeps resubmitting as new directories appear. Backlink outreach and content amplification live on Grow ($49/mo).

I hit the back button and lost my place

Browser back is supported. Typed-but-unsubmitted text is auto-saved locally so you won't lose work. If you're somehow on a page that looks empty after navigation, refresh, the server has your submitted answers.

I can't move forward, the page just sits there

Make sure all required fields on the current page have content. The continue button is disabled until required slots are filled. If everything looks filled and you're still stuck, refresh once, sometimes a transient state mismatch resolves on reload.

I want to start over

From /onboard/start there's a “start over” link. Clearing your browser's local storage for envvoy.ai also resets you. Or open an incognito window, the session token is per-browser.

These docs cover what envvoy does and how to use it. For pricing, billing, marketplace specifics, beta-end behavior, and the “why” questions, the FAQ is the better page. For the architecture, how the interview, claim, and brand kit pipeline fit together, see how it works.