This is a submission for the Hacktoberfest Open-Source AI Challenge Week 2: Side Hustle
What I Built
One-liner: PromptPilot turns a rough, half-formed requirement into a complete, editable prompt — and keeps it in a library you can search and reuse.
I am a game 3D artist. In my AI-assisted work I kept hitting the same small friction: I usually know what I want, but my first description of it is rough — a sentence like "write a Snake game" — and then I spend several rounds rewriting the prompt instead of doing the actual work.
PromptPilot is a lightweight desktop tool I built for myself to remove that loop. You type a rough requirement, pick one of six optimization modes (Standard, Professional, Concise, Structured, Creative, Direct), an output language, and a model — and PromptPilot expands it into a structured, high-quality prompt, streamed back token by token. The result is split into the model's reasoning and the final prompt, and the final prompt stays editable: it is a draft you refine, not a black box you must accept.
Optimized prompts are not one-shot either. PromptPilot stores them in a local prompt library with categories, keyword search, filters, favorites and usage statistics, so a good prompt becomes an asset you reuse instead of a lucky accident you lose. It also works the other way around: upload an image and it can infer the prompt that would reproduce it, or use the image as context while optimizing a text prompt.
I first showed PromptPilot publicly in my half-year work review presentation, alongside the production tools I build for my team.
Who Would Pay for It
People who work with AI assistants every day — developers, designers, writers, indie makers — and lose time the same way I did: not on the work itself, but on rewriting the prompt before the work can start, and on re-deriving good prompts they know they wrote once but never saved.
The side-hustle plan: the current local app — optimization plus the prompt library — stays free. A paid Pro tier adds what a local single-user tool cannot do: cloud sync of the library across devices, team workspaces with shared prompt libraries, and collaboration features for small studios. A one-time purchase or a small subscription both fit a tool of this size. The point is that it saves a few minutes many times a day, which is exactly the kind of tool people quietly pay for.
Demo
PromptPilot comes in two forms. The mini window is summoned anywhere with a global hotkey for quick optimizations — type a rough requirement, pick a mode and an output language, and send. The full window is the complete workspace: a home page with live usage stats and a word cloud built from my optimized prompts, plus the prompt library, history, and settings.
Code
Source code: https://github.com/natinli/PromptPilot-public
How I Built It
PromptPilot is a Tauri 2 desktop app: a React 19 + TypeScript frontend (Vite 6, Tailwind CSS v4, shadcn/ui) with a Rust backend, and a local SQLite database (via rusqlite) that stores the prompt library on the user's own machine.
The AI layer is provider-based rather than tied to one vendor: it supports Google Gemini, any OpenAI-compatible endpoint, and CodeBuddy, and the user can switch providers and models. Streaming output is supported per provider. Because the expansion layer only depends on a provider interface, the model doing the work is a choice, not a commitment.
I built it the way I build my production tools at work: start from a repeated annoyance in my own workflow, keep the tool small enough to actually finish, and document how it is used and checked, so the method does not only live in my head.
Why Open Innovation Matters
The provider design is the point. PromptPilot does not bet on one vendor: because any OpenAI-compatible endpoint works, an open-weight model — served by a local runner or a hosted endpoint — can power the optimization just as a proprietary API can. That keeps the cost structure honest (a side hustle dies if every click depends on one provider's pricing), it lets users keep their prompts and library on their own machine, and it means the tool survives any single provider's policy change. Open models and open interfaces are what make a tiny tool like this viable at all.
Top comments (0)