ClownfishWhy Clownfish
Built for one-person companies
When you're the whole company you write the code and also watch the site, answer customers and track the numbers. Keep building; hand clownfish the other three.
Read the code once, get three
One pass over your code drafts the pages to test, the support knowledge base and the key pages to watch. No explaining your product to three separate tools.
One tool, lower cost
One dashboard and one deployment instead of three tools; the model costs money only when a visitor asks. The parts talk to each other: when a test finds a broken page, support tells visitors who ask about it.
Teams can use just one part
Tests, analytics and support switch on separately. Some teams run only the tests; others only support, answering customers from their knowledge base.
What it does
Tests
Uptime checks on the pages that matter, each with the sentence it should show. Mail and chat alerts on failure and recovery; your editor's assistant can trace a failure to the commits around it.
Support
A chat window from one <script> tag. The bot answers only from your knowledge base and hands off to you when it can't; reply from the dashboard or your team chat.
Analytics
Visits on key pages, conversations, handoffs, visitors' first questions and uptime, in a weekly report every Monday.
Team chat
The same support bot lives in Telegram and Slack; Teams, Discord, Feishu, DingTalk and WeCom get alerts. Team and customer groups stay apart.
Drafted in your editor
- 1
Add your site and repos
Frontend, backend, several apps: add them all. A requirements doc is optional.
- 2
Draft in your editor
The Claude Code plugin reads your code locally and drafts tests, the knowledge base and key pages. Private repos need no access grant; it runs on your own assistant.
- 3
Confirm piece by piece
Review each part before it goes live. A GitHub Action runs the tests on every commit.
Free, and cheap to run
- Free to use: sign up and start.
- Uptime checks, the chat window, chat bots and analytics cost no model spend; the model runs only when a visitor asks.
- Any model: OpenAI, Claude, Gemini, DeepSeek, Qwen, local Ollama and more.
- Nothing runs before domain verification; every outbound request is limited to public addresses.