AI Domain Finder

Describe your product in plain English. Get a shortlist of brandable, confirmed-available domains with a clear register / consider / monitor / skip recommendation per result.

Built for founders, indie hackers, agencies, product teams, and LLM agents (via MCP) who need to move from vague idea to a registerable shortlist in one run — without manually pasting names into a registrar search bar.

---

Who this is for

• Founders & solo builders naming a new startup, SaaS, marketplace, or app
• Agencies presenting naming options to clients across multiple verticals
• Product teams validating brand directions before design work begins
• LLM agents (via MCP) that need a structured, decision-ready domain pipeline as a tool call
• Growth teams sourcing fresh microsite or campaign domains

---

What it actually does

Most domain tools assume you already know the name. They make you type words in and click search. This Actor flips that: you describe your idea, and it returns 5-10 names that passed every quality gate and are confirmed available right now via authoritative sources.

The full pipeline (11 quality gates)

1. AI classifies your idea into one of 11 business types
     → marketplace, saas-b2b, saas-b2c, dev-tool, ai-product,
       consumer-app, content-site, agency, portfolio, ecommerce

2. AI generates candidates biased to your idea-type
     → 40-candidate pool, distributed across naming strategies
       (portmanteau / compound / made-up / metaphor / verb-first / minimal)
     → Targets best-fit TLDs first (e.g. .com for marketplace, .dev for dev-tool)

3. Hard substring denylist (zero-AI-cost gate)
     → blocks names with scatological/sexual/slur substrings

4. Awkward-suffix filter (zero-AI-cost gate)
     → drops -dept, -corp, -ltd, -gov, -official, etc. for non-agency types

5. DNS validity check
     → enforces RFC 1035 (lowercase alphanumeric + internal hyphens, ≤63 chars)

6. RDAP + native whois availability check (FREE, no API key)
     → modern RDAP protocol + TCP whois on port 43
     → multi-source consensus: high/medium/low confidence per result
     → Domainr + Porkbun supported as optional enrichers (price + premium flag)

7. AI relevance scoring (0-10)
     → drops names below your relevance floor (default 5.0)

8. AI typo-resistance check (NEW!)
     → catches names that LOOK LIKE typos of real words
     → "dogstrid" reads as botched "stride" → flagged
     → "walkoos" reads as genuine invented brand → passes

9. AI multi-language safety check
     → flags negative connotations across EN/ES/FR/IT/DE/NL/PT
     → catches phonetic matches the substring list misses

10. AI brand-clash check
     → flags collisions with famous existing brands (Zoom, Google, Stripe, etc.)
     → severity: high/medium drops, low noted but kept

11. Final scoring + ranking
     → totalScore = relevance × 0.6 + memorability × 0.10 + tld-fit × 0.30
     → ranked output with `recommendedAction` decision signal

Retry loop until count is met

If the first batch doesn't yield enough names, the Actor runs up to 5 retry batches with growing pool sizes and seen-name exclusion. The final 1-2 retries enter rescue mode (relaxed relevance floor + skipped brand-clash) to honor your count request when filters are aggressive.

---

Key features

Feature	Detail
Plain-English input	Just describe your idea — no need to invent names yourself
11 quality gates	Substring, suffix, DNS, RDAP, whois, relevance, typo, safety, brand-clash, scoring, TLD-fit
Free authoritative availability	RDAP + native whois — no API key required for the default check
Optional API enrichers	Domainr (aftermarket signal) + Porkbun (price + premium flag)
Multi-AI provider	Use OpenAI or Anthropic; works with both
Idea-type classification	Auto-classifies your idea so TLDs and prompts adapt (marketplace gets .com priority, dev-tool gets .dev/.io, etc.)
Decision-ready output	recommendedAction: register / consider / monitor / skip
Per-result rationale	pickReason + risksAndCaveats explain every recommendation
Drop transparency	Every filtered-out name listed in droppedCandidates with reason
Per-TLD diagnostics	See exactly how many candidates were generated/available/returned per TLD
Token + cost tracking	Every run reports its own token usage + estimated USD spend
MCP-ready output	Single decision field per result, predictable schema, structured for LLM consumption

---

Input example

Minimum viable input — just idea and aiApiKey:

{
  "idea": "a marketplace for freelance dog walkers in Europe",
  "aiApiKey": "sk-ant-..."
}

Full input with all tuning options:

{
  "idea": "a marketplace for freelance dog walkers in Europe",
  "ideaType": "marketplace",
  "keywords": ["walk", "paw", "dog"],
  "tlds": [".com", ".io", ".co", ".ai", ".app"],
  "count": 10,
  "candidatePoolSize": 40,
  "style": "creative",
  "avoidWords": ["pup"],
  "maxLength": 12,
  "aiProvider": "openai",
  "aiApiKey": "sk-...",
  "domainrApiKey": "optional-domainr-key",
  "porkbunApiKey": "optional-porkbun-key",
  "porkbunSecretKey": "optional-porkbun-secret",
  "excludePremium": true,
  "relevanceFloor": 5,
  "runTypoCheck": true,
  "runSafetyCheck": true,
  "runBrandClashCheck": true
}

---

Output example

Each row in the dataset is one confirmed-available domain with a full decision package:

{
  "domain": "pawmile.com",
  "name": "pawmile",
  "tld": ".com",
  "available": true,
  "confidence": "high",
  "premium": false,
  "checkedVia": "rdap+whois",
  "ideaType": "marketplace",

  "relevanceScore": 8.8,
  "memorabilityScore": 8.5,
  "tldScore": 10,
  "totalScore": 9.1,

  "namingStrategy": "compound",
  "aiRationale": "Suggests a walking distance for dogs.",
  "pickReason": "Compound of two real words — Suggests a walking distance for dogs. .com is the strongest TLD for a marketplace. Directly evokes the concept.",
  "risksAndCaveats": [],
  "recommendedAction": "register",

  "registrarUrl": "https://dash.cloudflare.com/?to=/:account/domains/register/pawmile.com",
  "estimatedPrice": null,
  "generatedAt": "2026-05-14T18:27:34.961Z"
}

How to use the output

Field	What to do with it
recommendedAction = "register"	Score ≥ 8.0, high confidence, no critical risks. Buy now.
recommendedAction = "consider"	Score ≥ 7.0 but with caveats — read risksAndCaveats first
recommendedAction = "monitor"	Borderline. Worth watching if higher picks are taken.
recommendedAction = "skip"	Premium-priced or quality issues. Pass.
confidence = "high"	RDAP + whois both confirmed available
confidence = "medium"	Only one source decisive (often on .app/.dev — Google strips public whois)
confidence = "low"	Sources disagreed — verify manually before buying
premium = true	Registry holds at marked-up pricing ($100s–$1000s, not standard rate)
pickReason	One-sentence rationale combining naming strategy + TLD fit + relevance
risksAndCaveats	Empty = clean. Non-empty = read each line before deciding.
registrarUrl	Direct deep-link to Cloudflare Registrar (at-cost) or Porkbun if configured

---

OUTPUT.json — full run summary

The key-value store gets a complete run report:

{
  "input": { /* echoed input */ },
  "stats": {
    "generatedCount": 113,
    "filteredOutCount": 0,
    "checkedCount": 113,
    "availableCount": 14,
    "returnedCount": 10,
    "requestedCount": 10,
    "metRequestedCount": true,
    "retryBatchesRun": 5,
    "rescueModeUsed": true,
    "recommendedActions": { "register": 6, "consider": 2, "monitor": 2, "skip": 0 },
    "processingMs": 103456,
    "checkedVia": "rdap+whois",
    "scorer": {
      "inputCount": 14,
      "hardDenylistDrops": 0,
      "awkwardSuffixDrops": 0,
      "relevanceFloorDrops": 1,
      "typoDrops": 1,
      "safetyDrops": 0,
      "clashDrops": 5,
      "survived": 10
    },
    "bestFitTldsForIdea": [".com", ".co", ".market", ".shop", ".store"],
    "bestFitReturned": 7,
    "tldBreakdown": [
      { "tld": ".com", "requested": true, "generated": 113, "available": 10, "returned": 7 },
      { "tld": ".io",  "requested": true, "generated": 8,   "available": 4,  "returned": 2 },
      { "tld": ".co",  "requested": true, "generated": 54,  "available": 0,  "returned": 0, "note": "Candidates generated but none passed availability check" },
      { "tld": ".ai",  "requested": true, "generated": 0,   "available": 0,  "returned": 0, "note": "No candidates generated for this TLD — AI ignored or rejected it" }
    ],
    "tokenUsage": {
      "totalCalls": 28,
      "inputTokens": 14788,
      "outputTokens": 7727,
      "estimatedCostUsd": 0.114,
      "byCall": {
        "classify-idea":         { "callCount": 1, "inputTokens": 322,  "outputTokens": 8,    "estimatedCostUsd": 0.0009 },
        "generate-candidates":   { "callCount": 6, "inputTokens": 8273, "outputTokens": 7052, "estimatedCostUsd": 0.0912 },
        "score-relevance":       { "callCount": 6, "inputTokens": 1303, "outputTokens": 274,  "estimatedCostUsd": 0.0058 },
        "check-typo":            { "callCount": 6, "inputTokens": 2925, "outputTokens": 71,   "estimatedCostUsd": 0.0080 },
        "check-safety":          { "callCount": 5, "inputTokens": 1165, "outputTokens": 246,  "estimatedCostUsd": 0.0054 },
        "check-brand-clash":     { "callCount": 4, "inputTokens": 800,  "outputTokens": 95,   "estimatedCostUsd": 0.0030 }
      },
      "pricingAsOf": "2026-05"
    }
  },
  "droppedCandidates": [
    { "domain": "walkglow.com",  "gate": "relevance-floor", "reason": "relevance 4.5 below floor 5.0", "score": 4.5 },
    { "domain": "dogbyrd.com",   "gate": "typo-check",      "reason": "looks like typo of 'dog bird'" },
    { "domain": "walkzoom.com",  "gate": "brand-clash",     "reason": "medium clash with Zoom" },
    { "domain": "dogwhizz.com",  "gate": "safety-check",    "reason": "sounds like 'whiz', slang for urination" }
  ],
  "domains": [ /* full ranked results */ ]
}

---

Cost transparency

The Actor reports its own token spend per run. Typical numbers (using gpt-4o):

Scenario	AI calls	Estimated cost
Typical (no retries needed)	6	$0.022
Single retry triggered	12	$0.045
3 retries (full quality)	21	$0.084
Worst case (5 retries + rescue mode)	28	$0.13

Disable optional AI checks via input flags (runTypoCheck, runSafetyCheck, runBrandClashCheck) for cheap mode — saves ~30-40%.

Anthropic Claude Sonnet runs ~30% more expensive than OpenAI gpt-4o for the same call pattern.

---

Input parameters

Required

Parameter	Type	Description
idea	string	Plain English description of your business / product / project
aiApiKey	string (secret)	API key for the chosen AI provider

Common tuning

Parameter	Type	Default	Description
keywords	string[]	[]	Anchor words the AI should bias toward
tlds	string[]	[".com", ".io", ".co", ".ai", ".app"]	TLDs to generate + check
count	integer	10	Number of confirmed-available domains to return (1-50)
style	enum	creative	creative / professional / playful / minimal
maxLength	integer	15	Max characters in name part (4-30)
avoidWords	string[]	[]	Words/fragments to exclude from suggestions
aiProvider	enum	anthropic	anthropic or openai

Advanced quality controls

Parameter	Type	Default	Description
ideaType	enum	auto-classify	Skip AI classification by setting manually: marketplace / saas-b2b / saas-b2c / dev-tool / ai-product / consumer-app / content-site / agency / portfolio / ecommerce / unknown
relevanceFloor	integer	5	Drop candidates below this relevance score (0-10)
candidatePoolSize	integer	adaptive	Override the per-batch generation size (10-60)
excludePremium	boolean	true	Filter out registry-premium-priced domains
runTypoCheck	boolean	true	Enable AI typo-resistance check (~1 AI call)
runSafetyCheck	boolean	true	Enable AI multi-language safety check (~1 AI call)
runBrandClashCheck	boolean	true	Enable AI brand-collision check (~1 AI call)

Optional availability enrichers

Parameter	Type	Description
domainrApiKey	string (secret)	Adds aftermarket / marketed-for-sale signal
porkbunApiKey	string (secret)	Adds price quotes + premium flag
porkbunSecretKey	string (secret)	Required with Porkbun API key

The default RDAP + whois availability check works without any of these keys.

---

Using via MCP (LLM agents)

This Actor is designed for LLM agent consumption via the Apify MCP server. The output schema is deliberately flat + decision-ready:

For LLM consumers, the key fields are:

• recommendedAction — single signal (register / consider / monitor / skip)
• risksAndCaveats — array of caveat strings (empty = clean)
• pickReason — pre-synthesized 1-line explanation
• confidence — availability check confidence

Suggested LLM filter:

// Pick the strongest result with no caveats
const winner = results.find(r =>
    r.recommendedAction === "register"
    && r.risksAndCaveats.length === 0
    && r.confidence === "high"
);

If the LLM consumer needs full reasoning, the OUTPUT.json blob includes all stats, drops, and TLD breakdown.

---

What the Actor will NEVER do

• Return a name that contains vulgar / scatological / sexual / slur substrings
• Return a name that "looks like a typo" of a real English word (e.g. dogstrid)
• Return a name with high or medium severity collision with a famous brand
• Return a name where the AI's safety check flags negative connotations across EN/ES/FR/IT/DE/NL/PT
• Return a name flagged as registry-premium (unless you explicitly set excludePremium: false)
• Return a name marked as unavailable by RDAP+whois (fail-closed on disagreement)

If a result makes it through these gates, it's safe to put on a business card.

---

What it WON'T guarantee

• Trademark clearance. The brand-clash check catches well-known brands but doesn't replace USPTO / EUIPO / local TM searches. Always verify TM before launching.
• Cultural appropriateness in every market. Safety covers 7 major languages. If you're launching in Mandarin/Hindi/Arabic markets, run a native-speaker review.
• Pronunciation in every accent. Memorability scoring is English-phonetic. Multi-region brands may need spoken-aloud testing.
• Pricing on premium domains. Premium domains are filtered by default; if you set excludePremium: false, pricing comes from Porkbun (if configured) or is null.

---

Troubleshooting

Problem	Likely cause	Fix
metRequestedCount: false in OUTPUT.json	Your idea + TLD list is saturated, or filters too strict	Lower relevanceFloor, add more TLDs, broaden style, relax maxLength
All results are .io / .app not .com	All .com names for your idea are registered	Try a more specific idea description, or adjust keywords
confidence: medium on .app/.dev results	Google strips public whois for these TLDs	RDAP is still authoritative. Add a Porkbun key for second-source verification.
High clashDrops count	AI keeps generating brand-adjacent names	Add famous-brand stems to avoidWords
Cost higher than expected	Multiple retries fired (saturated namespace)	Check stats.retryBatchesRun and rescueModeUsed. Adjust relevanceFloor to keep typical runs cheap.
0 candidates generated for a TLD	AI judged TLD a poor fit for your idea type	This is intentional. Remove the TLD from input, or accept the breakdown's transparency.

---

License

MIT