# Thingsmith
**A little forge for naming your next big thing.**
A free, open-source AI naming skill for Claude Code, Codex and Gemini CLI, with domain, handle and pronunciation checks.
Finding a name you like is one job. Finding out whether you can use it is another.
Thingsmith is a free agent skill for naming products, brands and companies across markets. It develops candidates and gathers domain, handle and language observations behind a usable shortlist. You get the reasoning, the rejection log and the checks that still need a human.
I built it because I kept needing international names and getting stuck in the same loop: a promising idea, a taken domain, another idea, an awkward meaning somewhere else.
## Try it
In Claude Code:
```text
/thingsmith a tool that helps small studios find public tenders, for the UK and Poland
```
Already have a name? Ask directly:
```text
/thingsmith check thingsmith: .com and .io, GitHub and npm; markets UK and Poland
```
That goes straight to investigation. A naming request starts with a brief and develops candidates. A comparison request can work with your existing shortlist.
The skill is free under [MIT](LICENSE). Your agent's model usage, domain purchases and professional checks may cost money.
## What comes back
A report records the brief and candidate derivations, with creative comparisons and a rejection log. A neutral HTML sheet shows finalists in the same everyday situations, alongside competitors. You choose what to pursue.
The process is: brief → competitor map → territories → generate → screen → compare → investigate → present.
Compact scope targets 40–60 distinct ideas. Full scope targets 150+ using differently briefed generators where supported. These are effort budgets, not quality guarantees. The report states what actually ran, including models, isolation and skipped checks.
## Check the evidence
The scripts inspect registration and public-platform records. They distinguish what was observed from whether you can claim something:
| Result | Meaning |
|---|---|
| `taken` | A registration or public record was found |
| `NO_RECORD_FOUND` | This lookup found no record; registration or claiming remains unconfirmed |
| `UNVERIFIED` | Ambiguous response, hidden-holder possibility, blocked request or failed lookup |
| `UNSUPPORTED` | This tool cannot perform the check |
| `INVALID` | Input doesn't fit a supported platform rule |
| `MANUAL` | A human/provider check remains open |
A missing DNS record never proves a domain can be registered. A missing public profile never guarantees a handle can be claimed. Reservations, premium pricing and eligibility still matter. Evidence exports record timestamps and sources; old results need re-checking.
Exit 0 means the requested automated checks completed. It doesn't mean the name is suitable or available. Manual work remains listed separately. Exit 1 means requested observations are unresolved, unsupported or invalid; exit 2 means a CLI usage error. Pronunciation tools use exit 0 for completed machine checks even when a transcript differs.
## Say it in the markets that matter
An optional speech tool reads candidates using language-specific voices, then transcribes the speech without supplying the spelling. It flags possible spelling or pronunciation trouble. A machine mismatch is a reason to investigate; ordinary local accents don't have to sound identical.
The included [native-speaker form](assets/native-speaker-check.md) helps you get real feedback. A machine PASS isn't human validation. Preliminary trademark research also leaves professional clearance outstanding.
## Between a brainstorm and a naming agency
Asking ChatGPT for names is a useful start. Thingsmith adds a repeatable process, scripts for investigating the shortlist and a record of what remains unknown.
At the other end, Lexicon publishes an average project fee of **US$75,000–$150,000** for its professional naming engagement. That's a price anchor, not a claim that this skill provides equivalent work. [Lexicon's process and pricing](https://www.lexiconbranding.com/brand-naming-process/) (checked 22 October 2026).
Thingsmith is for the work between those options: moving beyond suggestions without buying a full agency engagement.
## Install
Requires Python 3.9+ and an agent that can read instructions and run local scripts. No API key is required by the core checkers. They make outbound requests to public registries and platforms; your agent host has its own model/data settings. Don't put private naming briefs in public issues.
**Claude Code — Windows PowerShell:**
```powershell
git clone https://github.com/samshennan/thingsmith.git "$HOME\.claude\skills\thingsmith"
```
**Claude Code — macOS/Linux:**
```sh
git clone https://github.com/samshennan/thingsmith.git ~/.claude/skills/thingsmith
```
If that destination already exists, update the existing clone rather than cloning over it.
**Codex / Gemini CLI:** clone to a location of your choice and explicitly load its `AGENTS.md` / `GEMINI.md` as task instructions. These are generated editions of the same skill, not a claim of automatic discovery by every host. Keep your host's permissions and approval settings.
Optional dependencies unlock additional capabilities:
| Install | Capability |
|---|---|
| `python -m pip install dnspython` | Fast DNS evidence that a domain is taken |
| `python -m pip install certifi` | Certificate bundle for handle requests |
| [eSpeak NG](https://github.com/espeak-ng/espeak-ng/releases) | Language-specific IPA and fallback speech |
| `python -m pip install faster-whisper` | Synthetic speech-to-text test; downloads a model on first use |
Windows SAPI voices are used when available; eSpeak is the fallback. Missing voices/backends remain explicit. Use `python3` where your system requires it.
## Run the scripts yourself
From the clone:
```text
python scripts/check_domains.py --names thingsmith --tlds com io --json domains.json
python scripts/check_handles.py --names thingsmith --only github npm pypi --json handles.json
python scripts/check_handles.py --list
python scripts/say_it.py --names thingsmith --langs en pl --readings-only
python scripts/context_sheet.py --names exampleone exampletwo --descriptor "your product description" --competitors Example --shuffle --sender "Your Name" --out shortlist.html
```
These are commands to investigate, not assertions about current availability. Check only the namespaces your project needs. Respect provider restrictions and rate limits; unsupported endpoints can change over time.
## Common questions
### How do I use AI to name a product or company?
Give Thingsmith the product, audience and target markets. It develops naming directions and candidates, then gathers observations behind a shortlist. Start with `/thingsmith name a product that …`. Existing names can go straight to checking.
### Can it check domains and social media handles?
It checks domain registration records and supported public handle/package namespaces. `NO_RECORD_FOUND` means a lookup found no record, not that registration is guaranteed. Some social platforms and hidden accounts require manual checks. Run `python scripts/check_handles.py --list` for coverage.
### Can it find an international brand name?
It screens candidates in the languages and markets you specify, using research and optional synthetic pronunciation tests. Native-speaker feedback is still needed for cultural meaning and practical pronunciation.
### Do I need API keys or a paid naming service?
The core Python checkers need no API keys. The skill is MIT-licensed and free; your agent host may charge for model use. Domains and professional legal/language review can cost money separately. The scripts don't read your API keys or send them with lookups; the agent host retains its own permissions and data settings.
### Does this replace trademark clearance?
No. Domain and handle observations don't establish trademark rights. Thingsmith can organise preliminary conflict research and a brief for an appropriate professional.
## Project status and contributing
Early release. Thingsmith itself remains a working name; its trademark and native-speaker checks are open. Historical domain/handle checks don't settle current availability.
The method lives in [SKILL.md](SKILL.md), stage guidance in `references/`, tools in `scripts/`, and templates in `assets/`. External methods are credited in [sources](references/09-sources.md); the agencies haven't endorsed this project.
Run offline checks with:
```text
python -B -m unittest discover -s tests -v
python -B scripts/sync_adapters.py --check
```
See [contributing](CONTRIBUTING.md), the [media kit](docs/media-kit.md) and the [Thingsmith field note](https://www.samshennan.com/tech/blog/thingsmith/) ([article source](docs/tech-post.md)). Edit `SKILL.md`, then regenerate the shipped adapters with `python -B scripts/sync_adapters.py`.
## Want help finishing?
[Moonacle](https://moonacle.com) offers a done-for-you route. Contact [contact@moonacle.com](mailto:contact@moonacle.com) with what you're naming and where it will sell to discuss scope.
