// the find
GeekyWizKid/writing-helper
A Next.js-based AI writing assistant supporting multiple LLM APIs (OpenAI, Claude, Gemini, etc.) with rich style customization features to help content creators improve quality and efficiency.
A Next.js writing assistant that wraps multiple LLM APIs with a style prompt editor, targeting Chinese-language content creators (WeChat articles, etc.). The main differentiator is the 'AI humanizer' feature that tries to defeat AI detectors by optimizing perplexity and burstiness statistics. Narrow audience: primarily useful if you're writing in Chinese and need WeChat-formatted output.
The multi-provider API abstraction (OpenAI, Grok, Ollama, DeepSeek, custom endpoints) is practical — switching models doesn't require code changes. The style editor's multi-dimensional prompt decomposition (tone, structure, narrative POV, emotional register as separate controls) is more thoughtful than a single free-text prompt box. The streaming proxy route correctly handles long-running generation with a 10-minute timeout, avoiding the default Vercel function limit that kills most naive implementations. MIT licensed with a reasonably clean Next.js 14 App Router structure.
The AI humanizer feature — designed to defeat GPTZero and similar detectors — is the kind of thing that will get academic users in trouble and is ethically questionable to ship as a first-class feature. API keys are configured client-side per the UI screenshots, which means they're sent to the proxy route every request rather than stored server-side; fine for personal use but a footgun for anyone deploying this for others. No persistence layer at all: no saved drafts, no history, no user accounts — everything lives in component state and is gone on refresh. The README's roadmap lists Claude and Gemini support as future work, but the description already claims them as supported, which is misleading.