Ayfri/Deep-Research

Ayfri/Deep-Research is a compact, approachable SvelteKit implementation of a Perplexity/OpenAI research chat app, and the README does a good job showing the product intent quickly: regular chat, a dedicated Deep Research mode, configurable OpenAI analysis models, Perplexity-backed research passes, markdown rendering, citations, conversation history, dark mode, and UI/API-key configuration. The code layout also looks easy to orient in, with src/routes/api/chat, src/routes/api/deep-research, and src/routes/api/name separated from reusable components like ResearchSettings, ResearchSteps, ModelSelector, MessageContent, and SettingsModal. For a small project with 44 commits, 11 stars, and 1 fork, that structure makes it unusually readable for someone who wants to learn how an AI research workflow is wired into a Svelte app.

The strongest part is that the project is not just a thin prompt wrapper. The README explains a real multi-step flow: analyze the user’s question, break it into sub-questions, research each aspect with Perplexity, organize phases, synthesize a final answer, and expose progress/timing details. Recent commits also suggest practical iteration, including streaming Perplexity responses and centralizing model definitions, which are exactly the kinds of changes that make this sort of app feel better to use and easier to maintain.

The main things I would tighten before recommending it more broadly are trust and onboarding details. GitHub detects the repo as GPL-3.0, but the README badge and license section say MIT, so the license should be reconciled immediately. The README also says browser-stored API keys are “stored securely” in localStorage; I would phrase that more carefully and explain the tradeoff, because localStorage is convenient but not a secret vault. There also does not appear to be a GitHub Actions workflow, and the package scripts rely on svelte-check rather than a visible automated CI gate. Adding a basic CI workflow for install, pnpm check, and pnpm build, plus a small test or two around the research request helpers, would make outside contributors more confident. A short architecture note explaining where prompts, API calls, streaming, and client state live would also help adopters modify the project without spelunking through the UI first.