DomainForge 🛠️

Find brandable domain names that are actually available — end to end.

Squatters have hoovered up the obvious .com and .ai names. DomainForge brute-forces creativity instead: it takes a vibe (a brief and/or one or two seed words), expands it into thousands of candidates, scores them for brandability, checks real availability across the TLDs you care about, and hands you a ranked shortlist.

It works with zero configuration (press Enter through everything) and with zero API keys (pure-algorithmic). Plug in your own LLM — Claude, OpenAI, Gemini, or OpenRouter — for a smarter seed pool and a curated shortlist with rationales.

This grew out of a real hunt that turned up new domains for me. The fragmented scripts behind those finds are now one clean tool.

---

How it works

brief + base words + categories + knobs
   │
   ├─ curated corpus  (~75 categories, 2,600+ words: mythology, science, world
   │                   languages, food, water bodies, animals, plants, mountains,
   │                   neighborhoods, pop culture, constellations, alchemy, coinages…)
   ├─ LLM seed-gen + invented coinages         (optional — your model)
   ├─ Datamuse word graph  (semantic neighbours, synonyms, triggers, spelling)
   ├─ affix variants  (get-, -ly, -ify, vowel endings…)
   └─ two-word blends  (multiple + verse → multiverse…)
        │
        ▼  score: length · syllables · vowel-ending · pronounceability · rarity
        ▼  availability: instantdomainsearch over your TLDs (threaded, retrying)
        ▼  rank: LLM-curated shortlist with rationale (or score-ranked fallback)
        │
        ▼  candidates.csv  ·  available.csv  ·  shortlist.csv + shortlist.md

No API keys are required for generation (Datamuse) or availability checks (instantdomainsearch) — both are free, public services.

---

Install

As a CLI (gives you a domainforge command anywhere):

cd domainforge
pipx install ".[recommended]"        # isolated, recommended
# or:  pip install ".[recommended]"

Then just run domainforge. Add LLM support with extras:

pipx install ".[recommended,llm]"    # all providers
# or pick one:  ".[recommended,claude]" / "[...,openai]" / "[...,gemini]"

Or run without installing:

pip install -r requirements.txt      # core + recommended (wordfreq, certifi)
python3 -m domainforge               # same tool, module form

Add a key (optional) by copying .env.example to .env and filling in one provider, or just export ANTHROPIC_API_KEY=... in your shell.

python vs python3: modern macOS/Linux ship only python3 (there's no bare python). The examples below use the installed domainforge command so it just works everywhere. Didn't install? Swap in python3 -m domainforge (Windows: py -m domainforge).

---

Quickstart

Guided wizard (Enter accepts every default):

domainforge

One-liner, no LLM — blend two seed words and check .com/.ai:

domainforge --base "rune,forge" --check 200 -y

Full power with a brief + extra TLDs:

domainforge \
  --brief "an autonomous web agent that does multi-step workflows for you" \
  --base "multiple" \
  --tlds com,ai,io,co \
  --count 10000 --check 3000 \
  --shortlist 30

Defaults are big on purpose: generate 10,000 candidates and availability-check the top 3,000. Generation is fast and free; checking is the slow part (~10–20 min at the default polite concurrency). Lower --check for a quick pass, or raise it for an exhaustive sweep.

Outputs land in domainforge_out/:

File	What it is
shortlist.md / shortlist.csv	Start here. Top available domains, ranked, with reasons.
available.csv	Every name with ≥1 available TLD.
candidates.csv	The full scored pool with per-TLD status.

See examples/ for committed sample output.

Presets

Two shortcuts set the volume knobs for you (explicit flags still override):

domainforge --fast      --base "rune,forge"   # quick scan: count 2500, check 500, ~2-4 min
domainforge --thorough  --brief "..."          # deep sweep: count 15000, check 6000

---

Options

Flag	Default	Meaning
--brief	—	Company/vibe description; powers LLM seed-gen & ranking.
--base	—	One or two seed words (comma/space separated).
--categories	all	Curated categories to draw from (--list-categories to see them).
--tlds	com,ai	TLDs to check (com,ai,io,co,xyz,app,dev,so,…).
--count	10000	Target size of the candidate pool (generation stops early once reached).
--check	3000	Max names to availability-check, top-ranked first (cost/time cap).
--shortlist	25	Size of the final ranked shortlist.
--provider	auto	auto / claude / openai / gemini / openrouter / none.
--model	—	Override the provider's default model.
--fast / --thorough	—	Volume presets (quick scan / deep sweep).
--no-affixes, --no-blends	off	Turn off a combination strategy.
--workers	5	Availability-check concurrency (keep modest — be polite).
--outdir	domainforge_out	Output directory.
-y, --non-interactive	off	Skip the wizard; use defaults + flags + config.
--save-config	off	Persist resolved settings to domainforge.config.json.

Everything is also configurable via domainforge.config.json (copy the .example). Resolution order: built-in defaults → config file → flags/wizard.

---

Bring your own model

DomainForge auto-detects the first API key it finds (preference order: Claude → OpenAI → Gemini → OpenRouter) and uses it for three things:

1. Seed generation — turns your brief into on-theme words across many domains of knowledge most lists miss.
2. Coinage — invents brandable made-up words tuned to your brief.
3. Ranking — curates the available list into a shortlist with one-line rationales.

Provider	Env key	Default model	Override
Claude	ANTHROPIC_API_KEY	claude-opus-4-8	--model
OpenAI	OPENAI_API_KEY	OPENAI_MODEL (best-effort)	env / --model
OpenRouter	OPENROUTER_API_KEY	OPENROUTER_MODEL	env / --model
Gemini	GEMINI_API_KEY	GEMINI_MODEL	env / --model

The Claude path uses the official SDK with adaptive thinking. The non-Claude default model IDs are best-effort — if one 404s, set the matching *_MODEL env var (or --model) to your account's current model.

No key? No problem. Every step that needs an LLM is skipped and the pipeline runs fully algorithmically; the shortlist is ranked by the brandability score instead.

---

Notes & caveats

• Availability ≠ purchased. Results reflect instantdomainsearch's DNS/registry view. Always confirm at your registrar before buying.
• DomainForge does not register domains — it finds candidates.
• Datamuse and instantdomainsearch are free, rate-limited public services. Keep --workers modest; the checker already backs off on 429/503.
• wordfreq is optional but recommended — it powers the rarity tier that separates registrable gems from always-taken generic words.

Project layout

domainforge/
  seeds.py         curated corpus + category selection
  generate.py      Datamuse expansion, affixes, blends
  score.py         brandability score + frequency tier
  availability.py  instantdomainsearch checker (any TLD)
  llm.py           Claude / OpenAI / Gemini / OpenRouter abstraction
  config.py        layered config + .env loader
  pipeline.py      orchestration
  output.py        CSV / Markdown writers
  cli.py           flags + interactive wizard

Share your finds

Found a gem? Post it. Screenshot your shortlist.md, tag the domain you grabbed, and link back so other founders can dig too. The best names are the ones nobody else thought to generate — show people what the long tail looks like.

Contributing

PRs welcome — especially new seed categories (the more obscure, the better), extra TLDs, and additional LLM providers. Open an issue with ideas, or send a PR adding a category to domainforge/seeds.py.

---

MIT-licensed (see LICENSE). Built for founders, by founders.