FakeAPI — A Mutable, Isolated REST Sandbox for Testing and Prototyping

#devchallenge #githubchallenge

GitHub “Finish-Up-A-Thon” Challenge Submission

This is a submission for the GitHub Finish-Up-A-Thon Challenge

What I Built

FakeAPI is a hosted REST API for testing and prototyping — real in-memory CRUD, no database, no auth. You create a workspace and get your own isolated copy of seed data with full CRUD, filtering, sorting, and pagination across Tasks, Users, and Projects.

There are other tools in this space, but none hit the exact combination I needed:

	JSONPlaceholder	ReqRes	MockAPI	FakeAPI
No signup	✅	✅	❌	✅
Real CRUD (data actually changes)	❌	❌	✅	✅
Session isolation (your data is yours)	❌	❌	❌	✅
Pre-loaded seed data	✅	✅	❌	✅
Reset to seed state on demand	❌	❌	❌	✅
Multiple resource types	✅	❌	✅	✅

JSONPlaceholder acknowledges your writes but doesn't store them. ReqRes returns hardcoded responses. MockAPI requires an account and you define the schema from scratch. FakeAPI is the only one that gives you real, mutable, isolated data without any setup.

Demo

Live: https://fakeapi-kynw.onrender.com (free tier — first request may take ~1 min to wake the server)

# Create your isolated workspace
curl -X POST https://fakeapi-kynw.onrender.com/api/workspaces/

# Browse pre-loaded tasks, users, projects
curl "https://fakeapi-kynw.onrender.com/api/workspaces/{id}/tasks/?status=pending&sort_by=-created_at"

# Reset to seed state any time
curl -X POST https://fakeapi-kynw.onrender.com/api/workspaces/{id}/reset

Interactive docs: https://fakeapi-kynw.onrender.com/docs

The Comeback Story

I built a rough first version in February 2025, deployed it to Render, and forgot about it for 15 months.

Before — a single /tasks endpoint backed by statically generated fake data using faker. No real storage, no isolation, no CRUD. Every caller saw the same global state and couldn't change anything that persisted. The README was one line. There were no tests, no deployment worth sharing beyond a proof of concept.

The core problem was that I'd solved the wrong thing: I had endpoints that returned data, but not an API you could actually use to build or test something. If you created a task, it was gone on the next request. If two people hit it at the same time, they'd interfere with each other. It was a demo, not a tool.

After — the architecture changed completely. Workspaces became the central concept: each caller gets their own in-memory copy of all seed data, isolated from everyone else. You can create, update, delete — and none of it bleeds into another session. A background loop cleans up expired workspaces hourly, each one lives 24 hours and can be extended up to three times.

On top of that: three full resources (Tasks, Users, Projects) with filtering, sorting, and pagination; nested routes like GET /users/{id}/tasks; rate limiting; a full test suite with a coverage gate enforced in CI; a GitHub Actions pipeline; and a landing page.

My Experience with GitHub Copilot

My background is backend. I'm comfortable with Python, APIs, and data modeling — but when it came to building the landing page, I was starting from scratch.

The design I had in mind was a terminal-style interface: monospaced font, a fake browser bar with colored dots, dark mode toggle that persists across sessions. I knew what I wanted visually, but I didn't know CSS well enough to build it confidently. Copilot bridged that gap. I'd describe what I wanted — "a sticky header with a theme toggle that switches between light and dark using CSS variables and saves the preference to localStorage" — and it would give me something that worked. I'd tweak, ask again, and iterate.

What surprised me was how much I actually learned in the process. Because Copilot explained the patterns it used — CSS custom properties, the data-theme attribute approach, the FOUC prevention trick with an inline script before the CSS loads — I came out understanding the code, not just having it. That felt like the right way to use the tool.

If you're building a frontend, testing an HTTP client, or just need a realistic API without the setup overhead, give it a try.

Repository: github.com/cristianrubioa/fakeapi