AIToday
Large Language ModelsAI Coding AssistantsZenn AI/MLPublished: Sep 28, 2026, 13:00 JST

Developer builds Jev Bookmarks to start Jev from Chrome history

Developer builds Jev Bookmarks to start Jev from Chrome history

3 Key Points

  1. What happened

    While building an accounting app called Books tied to マネーフォワード, a developer built Jev Bookmarks. It searches a project's saved pages first; if none fit, it gathers candidates from Chrome history and lets Jev choose where to start, then Jev-ultrafast operates the browser.

  2. Why it matters

    The tool targets a real friction point: repeatedly opening the same pages, such as fixed-asset ledgers and tax filing screens, during development. Saving useful pages per project means similar later requests can start from an already-known place, cutting the effort of hunting for a familiar starting point.

  3. What to watch

    The developer noted no recorded comparison of speed in seconds, so the benefit is felt rather than measured. If neither history nor saved pages offers a suitable entry, Jev Bookmarks returns 'no entry,' and the AI must fall back to Jev's browser search.

WHO IT HITSDevelopers and AI agent builders who repeatedly drive a browser to the same internal tools—such as accounting dashboards and tax filing screens—may benefit from caching known starting points, though the body offers no measured speed comparison.

Not sure about something? Ask the AI

Questions and answers are published on this page.

Summaries like this, in your inbox every morning.

Context & Analysis

The developer's starting point was testing Jev-ultrafast and Jev's Computer Use after seeing them shared on X. The browser felt noticeably fast, which raised the question of where that speed actually helps. Two cases emerged: unfamiliar sites where the AI must explore, and known destinations the AI is asked to open repeatedly during development.

Building the Books accounting app tied to マネーフォワード made the second case concrete, as the AI frequently reopened screens like the fixed-asset ledger and tax filing pages. Since those URLs often sit in Chrome history, the developer had Jev choose a starting page rather than letting it operate blindly. The design sends only narrowed candidates to Jev, keeping full history and query strings out of the entry decision, and pages judged useful after an operation are saved per project for reuse.

The developer also noticed Jev can evaluate several viewpoints at once, as described in TypeSafe's documentation. The current Jev Bookmarks picks a single entry, but keeping three candidates for the next step may be possible. Whether this approach pays off hinges on how often saved pages match a new request, since no time-based measurement was recorded.

FAQ
What problem does Jev Bookmarks solve?
Jev's browser operation is fast, but if you already know the destination, opening its URL directly is faster. The tool lets Jev choose the starting page from a project's saved pages or Chrome history instead of searching.
What happens when no suitable starting page is found?
Jev Bookmarks returns 'no entry.' The developer says the AI can then use Jev's browser operation to search instead.
How does Jev Bookmarks handle Chrome history data?
It reads history on demand through a Chrome extension, narrows candidates, and passes page titles plus the host names and paths of URLs to Jev. Full history and query strings are not sent for the entry-selection decision.

Get the latest Large Language Models news every morning

For example, today's edition would include:

  • AWS CloudWatch Omni now generally available with 17 built-in evaluatorsSiliconANGLE AI · 1h ago
  • Google Vids gets Gemini Omni 1.1 Flash, 1080p videoAI Watch (Impress) · 1h ago
  • OpenAI paper: AI can't say "I don't know"Qiita 機械学習 · 1h ago

AI-summarized, only the topics you pick: one digest a day via Email, LINE, or Slack.

Free · 30 seconds with Google · unsubscribe anytimeWhat is AIToday? →

Ask AI

Ask AI anything about this article. The AI reads this article, earlier AIToday articles, and Wikipedia, and cites its sources. Q&As are published on this page for other readers too.

Questions and answers are published on this page.

Related Articles

Next articleOpenAI halts training of top models after agent slips past network limits