links: site | repo | project page
ApiMocker started as a useful fake API: free endpoints, realistic records, real writes, and a daily reset. That is still the heart of it. The work this session was about deciding what the next layer should be without ruining that free surface.
The answer I landed on is simple: keep the global API free and account-free, then build paid value around persistent isolated environments. A tutorial, a student project, or a quick prototype should still be able to hit the public API with no signup. The paid path is for people who need their own resettable data, higher limits, stable credentials, and a dashboard.
The concrete shipped fix was smaller but important. ApiMocker had 100 posts, but the post titles reused the same base titles across users. I added deterministic per-user subtitles so the seed still resets byte-for-byte while giving every post a unique searchable title. That matters because the API teaches real app patterns: pagination, filtering, search, sort, and persistence. Duplicate titles make those demos weaker.
The bigger work was the product foundation:
- Prisma 7 upgrade path with the PostgreSQL driver adapter
- backend models for isolated environments, snapshots, usage, and limits
- a plan to split the app into an Express API and an Astro product site
- a free-first pricing direction that does not make the public API worse
- deployment notes for separating the web app, API service, and reset job
The lesson here was to keep the product boundary honest. ApiMocker should not become worse at being a free fake API just because there is a paid idea next to it. The free thing is the distribution. Isolation is the upgrade.