Project notes

Zhihu Threads

Choose Zhihu source excerpts and build a learning thread with follow-up questions and self-tests.

Project illustration of Lora and Mochi organizing source excerpts at their desk. — Project illustration, not a product screenshot.

The user chooses the evidence

The user selects Zhihu excerpts. AI organizes the selected material into a learning thread without silently changing the evidence boundary.

Clarify the question

Clarify the learning intent before retrieval. When the question is vague, AI helps narrow its scope.

Retrieve excerpts

Search returns summary-level excerpts from Zhihu answers or articles. Candidates retain their source type and link; they are not full article bodies.

Select sources

AI explains relevance and possible use. The user chooses the sources. Unselected excerpts stay outside the thread’s evidence boundary.

Follow up and self-test

AI organizes selected excerpts into learning nodes, open questions and self-tests. Follow-up answers remain grounded in those excerpts and retain evidence gaps.

This explains the workflow. It does not call a model or access your Zhihu account.

Workflow diagram. Clarify the question, retrieve excerpts, let the user select sources, then build a learning thread. Missing evidence stays unknown.

Implementation and limits

Stack: TanStack Start · TypeScript · Tailwind CSS · Effect · SQLite
Retrieval: Uses summary-level ContentText from the official API, not scraped or invented full article bodies.
Thread follow-up: Answers from the current thread’s excerpts. Missing evidence is marked unknown or evidence_gap.
Run records: The repository includes evaluation data, JSON traces and a local evaluation dashboard. Test counts are not quality scores; this page does not claim verified live-model performance.

These notes were checked against the repository on September 9, 2026. Current source code remains authoritative. View reviewed revision 6c34a81