Tolson64 / Kitlane-extension

English | 简体中文
# 🔖 Kitlane
> **Can't remember the name? You can still find the tool you saved.**
MIT · Chrome / Edge side panel · Chinese and English UI
A few hundred links, saved because they looked useful later. When later arrives, the name is gone. You scroll, search the web again, and open the wrong site.
Kitlane is for people who would rather not tidy. It leaves your folders alone. It reads the bookmarks you already have, tags each one by what it is for, and pulls up the right one when you type something like "a site that shrinks images."
## Try it
Everything below is local search. No model, and no API key.
| You type | What comes back |
|---|---|
| find skills | SkillsMP, anthropics/skills, skills.sh, Smithery |
| system design interview material | LeetCode, coding-interview-university, system-design-primer |
| English papers | arXiv, SSRN |
| financial reports | HKEXnews, SZSE, CNINFO, SEC EDGAR |
| a quant backtesting platform | RiceQuant, Portfolio Visualizer, JoinQuant, QuantConnect |
| macroeconomic data | FRED, Jin10, World Bank, National Bureau of Statistics |
| web design inspiration | Dribbble, Behance, Awwwards, Godly |
| an AI tool for slides | AiPPT, iSlide, Gamma |
| collecting payments from overseas | PingPong, LianLian Global, Payoneer, Wise |
The sample is 480 bookmarks the author collected. Treat the examples lightly. The numbers further down explain why.
For a tighter order, turn on Smart search and let a model you configure re-rank the results.
## Features
- **Classify with no API key.** On import, bookmarks are sorted using a built-in library of 920+ sites, URL patterns, your folder names, and page descriptions. No model is called.
- **Search by the task.** "Remove background" finds image tools. "Annual report" finds filings.
- **Collections follow the tags.** Sharper tags make sharper collections. A new collection also gets tag suggestions from its name.
- **Bring your own model, if you want one.** Any OpenAI-compatible API works, in the cloud or locally through Ollama or LM Studio. The Jev decision model is optional.
- **Private links stay on the device.** Internal tools, online documents, account consoles, cloud-drive shares, invite and share links, links that carry a key or credential, and adult content are never sent to a model and never fetched.
## How accurate is it
Share of bookmarks left unclassified when no model is configured (lower is better).
| Source | Before | Now |
|---|---|---|
| The author's bookmarks (57) | 78% | **32%** |
| Highly voted Hacker News and Lobsters links (405) | 75% | **45%**, or **29%** after reading page descriptions |
On search: 61 colloquial Chinese queries, 480 sample bookmarks. The right result lands in the local top 30 **97%** of the time, and in the top 5 **92%** of the time.
Those figures are generous. The author's own set is only 57 bookmarks, and both the sample library and the queries were written by the author, so the search scores are optimistic and only useful for comparing