When you learn with AI, you often skim a long answer, hit an unfamiliar concept, dig in, find another fuzzy term—and keep branching down along what you do not yet understand until each step is clear. Recursive Learn supports that workflow: questions become a traceable tree, each node can be marked learned vs. not, so you can deep-dive along one branch first, then jump back anywhere on the tree.
- Follow-up chains beat one-shot walls of text. The unit of learning is question → answer → follow-up, not a single dense summary.
- Mastery is explicit. Nodes can be tagged so you always see what still needs work.
- Structure survives revisits. Under one topic, threads form an exploration tree (rooted learning sessions)—you are less likely to get lost than in endless flat chat transcripts.
| Capability | Description |
|---|---|
| Start from a topic | On the home page, enter a topic to create a root node; the model answers with optional web search |
| Drill down | Ask a follow-up where you are stuck; create a child node so the thread stays structured, or use Just ask for a quick sidebar without branching immediately |
| Sessions & full tree | Sessions lists each root topic; open one session to browse the full tree, jump between nodes |
| Search | Search finds nodes across sessions by keyword and opens them |
| Local-first data | Learning data stays in IndexedDB on your machine; Settings lets you export / import JSON backups |
| Model & search | Configure provider and API keys under Settings; optional web search (e.g. Tavily, Brave, Exa—keys stored in the browser). Requests can route through your deployed app server before reaching vendors (environment variables supported on the server) |
-
Install and run (Node.js required):
npm install npm run dev
Open the app in your browser (usually
http://localhost:3000). -
Configure keys. Open Settings, pick an LLM provider and enter an API Key (or rely on server-side env). Add search keys only if you need live web retrieval. Save.
-
Begin a thread. Enter a topic and start learning; read the answer and ask the next concrete question beneath it. Branch with create child, or use Just ask for a lightweight Q&A without a new branch yet.
-
Mark mastery. Mark nodes as mastered or still open so next steps stay obvious.
-
Jump between themes. Use Sessions to list roots, open a tree view, create new roots, or delete a whole tree (confirm before destructive actions).
-
Back up data. Before switching browsers or clearing site data, use Settings → Data management to Export JSON, then Import elsewhere.
-
UI language. Toggle Chinese / English in the header; this does not force answer language—you control tone and wording in prompts.
- Next.js (App Router), React, TypeScript, Tailwind CSS
- Persistence: IndexedDB; logic split across
src/domainand components - Testing: Vitest, Playwright (
npm run test,npm run test:e2e)
Standard Next.js production run:
npm run build
npm startnpm start serves on port 3000 (see package.json).
For Cloudflare (preview / deploy):
npm run preview # OpenNext preview
npm run deploy # Deploy with OpenNext for Cloudflare- API keys and most learning context stay in your browser; when you prompt, keys may reach your deployment’s server before third-party APIs. Avoid saving keys on untrusted devices—clear them in Settings when done.
- If the server exposes shared keys via environment variables, behavior follows your deployment config.
Improvements to copy or onboarding are welcome via PRs—keep this README aligned with how “recursive learning” shows up in the product.