Back to Shelf

Roadmap — 04 · On the workbench

What the maker is thinking about.

Three sections, no quarter targets, no dates. The top is what has tools out today. The bottom is what shipped recently. The middle is the part that moves.

Filed in Verona · Updated 14 May 2026 · § Three items, hard cap

Currently on the workbench

Three items maximum. When a fourth arrives, one of these moves to shipped or drops down to set aside.

  1. Highlights.

    Text you select on a page saved alongside the article. The highlights export as nested block quotes in Obsidian and Logseq, and as toggle blocks under the source page in Notion. There is no separate highlights view planned — the highlight lives with the piece it came from. The maker is not promising a date. If this stalls, you'll read it here.

    on the workbench

  2. Local fuzzy search.

    Title, tags, excerpt. Indexed on save, queried in the popup. No embeddings, no infrastructure — just a fast pass over the local library that runs in the browser without the page knowing it exists.

    on the workbench

  3. Notion auto-sync.

    Background re-export of edited items for Pro readers. When you change a title, fix a tag, or move something into a collection, the Notion page updates without you re-running the export. Quiet, opt-in, throttled.

    on the workbench

Considering next

Items the maker is reading about and turning over. None of these have a build start. Some never will.

  1. Folder export sync.

    Using the File System Access API: you pick a folder once, Shelf writes Markdown files into it on every save, and your existing Dropbox or iCloud client syncs them across machines. Pro only. The maker likes this path because the cloud is yours, not ours — but the API has been quietly unstable across Chrome versions and we don't want to ship something that breaks on a quarterly update.

    considering — needs the File System API to prove out

  2. Reading Stack — public page.

    An opt-in, static, single-page render of the saves you've marked as favourite. No social graph, no comments, no metrics. The equivalent of taping a small reading list to your office door.

    considering

Set aside

Features the maker has considered and deliberately not started. Each carries a one-sentence reason so you don't have to guess.

  1. Cloud sync across devices.

    The single most-asked feature. The folder-export path above is our answer for now — your sync provider is already good at this, and a Shelf cloud would mean a Shelf server, which is the one thing the manifesto is built against. We may revisit in 2027 if File System Access proves out.

    set aside

  2. Mobile companion app.

    Shelf runs in Chrome because Chrome is where the reading happens. A mobile app would be a second codebase with a different attention model, and one indie maker cannot support both well. Markdown on disk and Notion pages are mobile-readable already.

    won't build — see /decisions

  3. Send-to-Kindle, ePUB export.

    Narrow audience. The format work is real, the testing burden is larger, and the PKM tribe is who Shelf is for. We don't want to dilute the positioning for an audience that has Pocket-Web, Instapaper, and Readwise already serving them well.

    set aside indefinitely

  4. Spaced highlights review (Anki SM-2).

    Spaced repetition is Readwise's home turf and they do it well. Competing on their best feature would mean building a second product inside Shelf, and the maker would rather link out to Readwise than ship a worse Readwise.

    won't build — see /decisions

  5. Semantic search via embeddings.

    Requires either a Shelf-run server to host the index, or a paid cloud API on every save — both violate the local-first positioning. The local fuzzy search above handles ninety percent of real intent and runs in the browser. We'll take the trade.

    set aside

  6. Podcast transcript via Whisper.

    Real per-minute compute cost and a tight coupling to the AI pillar. We'd rather lean on transcripts the publisher provides (YouTube has them, podcast hosts increasingly do) than run an inference bill on Shelf's behalf.

    set aside

  7. Twitter / X thread unroll and save.

    The X API is a graveyard and the rules change every quarter. Building on it means accepting that the feature can be broken by a TOS update we don't control. Not a fight Shelf wants.

    won't build — see /decisions

Recently shipped

The latest entries from the changelog. Full notes live there.

§ Changelog § Send feedback

※ End of edition ※