KiwiKit
A working mini SaaS demo built with AI coding tools. It turns a repetitive GTM scraping workflow into a shared API, dashboard, usage system, and playground your team can actually try.
This is intentionally capped for demos, workshops, and learning. It is not positioned as a high-volume production scraping plan.
The scraper is the example. The workflow is the lesson.
KiwiKit exists to show how Claude Code or Codex can help a non-traditional builder turn a painful repeated workflow into a small product. The AI helps build the tool, but the final workflow runs predictably on servers.
That distinction matters: the LLM is useful during the build. The live system should be boring, observable, and repeatable.
A small, real API surface for experimenting.
Use the dashboard, playground, and docs to test the same building blocks you would expect in a shared internal tool.
Submit URLs through the dashboard or API
Choose raw HTML, article, selectors, or extraction output
Generate demo API keys and test requests safely
Watch usage against the monthly 1,000-credit cap
Six pieces that turn a script into a team tool.
This is the framework from the session: if a workflow needs all six, it probably deserves a proper interface.
Input
URLs, prompts, files, or rows enter through a clean interface or API.
Process
The real work runs on normal server infrastructure, not inside a chat window.
Output
Every run returns predictable JSON that can move into another workflow.
Observability
Usage, failures, latency, and request history are visible to the team.
Storage
Shared state lives centrally so one person is not carrying the whole workflow.
Access Control
Users get keys, limits, and permissions instead of everyone sharing secrets.
Build with the LLM. Run without the LLM.
The demo is intentionally simple: a scraping API, keys, usage, docs, and a playground. But the architecture pattern is the interesting part. AI coding tools help assemble the system; once it works, the workflow should execute the same way every time.
{
"environment": "demo",
"monthly_credits": 1000,
"tool": "scraping_api",
"built_with": ["Claude Code", "Codex"],
"runtime": "server",
"lesson": "mini_saas_over_chat_workflow"
}