PromptForge.
Browser-based AI prompt engineering toolkit built for anyone writing complex LLM prompts. It ships with a clean, focused UI and handles the unglamorous parts: prompt versioning, comparison, and export.
01
Prompt complexity.
Users iterate on prompts dozens of times; without a good interface, edits are scattered across notes, Discord, and browser tabs..
02
Comparison under load.
Side-by-side prompt evaluation with real API calls has to stay responsive even when re-running expensive completions.
03
Export and sharing.
Users need their work portable — markdown, JSON, shareable links without lock-in.
04
API cost transparency.
Users need to see what they spending on API calls in real time, hidden costs breed distrust and churn.
05
Prompt versioning.
Tracking which version of a prompt produced which result matters, a naive approach creates clutter and confusion.
Component-driven architecture. React with TypeScript, Radix UI primitives for accessibility, Tailwind for styling. Each prompt iteration is a self-contained, copyable artifact.
State management with Context. prompt history, favorites, and current editor state live in Context, no external database for this prototype stage.
API integration. direct calls to Claude API (via the frontend, authenticated through environment) for real-time prompt evaluation.
Export pipeline. markdown, JSON, and shareable URL generation all client-side, no server-side persistence needed for the MVP.
Mobile
Prompt engineering is sometimes mobile work
tweaking from a phone between meetings. The UI is mobile-first, though the real work happens on desktop.
Live and usable today a focused prompt editor built around the friction points people actually encounter when writing LLM prompts: versioning, side-by-side comparison, and portable export.
ReactNextTypeScriptRadix UIError Boundaries
Have something like this in mind?
If this is the kind of work you need — performance, architecture, or a rebuild — a short call is the fastest way to find out whether I can help.