Ariadne – reason about your codebase on a walk, run, or commute
Voice-first coding agent when Cursor and GitHub Copilot already dominate this space.

The app ships a clear, science-framed workflow instead of the usual free-for-all whiteboard: you lock a problem, force phase separation (no premature judgment), and can swap LLM backends (Llama 3 or paid Gemini). It’s a useful, thoughtful wrapper around existing tools and models — smart UX choices like session autosave and constrained prompts show craftsmanship — but it’s not reinventing AI brainstorming so much as packaging best practices into a guided product.
Product builders, indie makers, designers, PMs and writers who want a structured, research-backed process for ideation
I work in analytics engineering (SQL, Python) and this is the first website I've put on the internet since MySpace. I built it while ironically trying to brainstorm ideas for side projects.
The origin: I asked Perplexity whether any app implemented the complete scientific creativity process — not just "AI brainstorming" but the actual validated sequence: preparation → divergent generation → incubation → convergent evaluation. It told me pieces exist (Notion AI for capture, various timer apps, scattered brainstorming tools) but nothing that enforces the full process as a single guided workflow. So I built it.
What it does:
The app is a 4-stage wizard based on Wallas' 1926 model, updated with modern neuroscience on Default Mode Network (DMN) and Executive Control Network (ECN) interactions:
Preparation — Frame the problem and define constraints. Constraints are intentional: research shows limits improve creative output by reducing the search space (Stokes, 2022). Optional: upgrade AI to Gemini 2.5 Flash with your own free key — or use the built-in Llama 3 backend with zero setup.
Generation — Judgment-free idea capture with AI augmentation (Generate / SCAMPER / Wildcard Constraint). No delete button. No scoring. UI designed to keep your ECN quiet and let the DMN run.
Incubation — A 10-minute countdown with rotating neuroscience cards on DMN activation, spreading activation theory, and why the most creative people alternate between focus and rest. Skip is available but discouraged. Mind-wandering during rest produces measurably better insights than continued focus (Baird et al., 2012).
Verification — Weighted criteria scoring, AI auto-scoring, AI refinement, and AI next steps generation. Tab-based mobile UI so you can switch between the ranked idea list and scoring panel without scrolling. Export full session as .txt.
Tech stack:
React + TypeScript + Vite → Cloudflare Pages (static, no server)
Cloudflare Worker → Cloudflare Workers AI (Llama 3 8B, free tier) Optional BYOK: Gemini 2.5 Flash via direct browser API — on any error, silently falls back to Llama 3
localStorage only — no database, no auth, no backend
Built almost entirely with vibe coding (Cline + Claude) since I don't write React day-to-day
Rough edges:
No cloud sync — sessions live in your browser only No accounts, no sharing, no analytics
Link: https://creativeflow.pages.dev/
Feedback welcome, especially if you find this useful or have another method for coming up with ideas.
Voice-first coding agent when Cursor and GitHub Copilot already dominate this space.
Seven stage gates with automated reviewers keep AI product work verifiable.
Comprehensive glossary and mental models, but it's a static blog post—not a product or tool.
Malicious OpenClaw skill scanner, but the market for hardening OpenClaw specifically is tiny.
Wraps a known method in chat—smart positioning, but five-minute utility beats lasting adoption.
Clean 3D quantum viz, but PhET simulations already cover this ground without AI.