Recall Ship Log, Reviewed: August 2026

What we read, and what this series is

We read Recall's public changelog at feedback.recall.it/changelog on 4 September 2026, and the newest entry on the board was dated 19 August 2026. This is the first in a recurring series in which we read a competitor's shipped releases and ask what each item cost to build and who it was for.

The method, stated up front so you can discount it appropriately. We read the public changelog and the public developer documentation, and nothing else. We do not have Recall's backlog, their telemetry, or any private word from their team. We are a competitor — BrainTube sells into the same category — and we would rather say so than dress this up as neutral analysis. What we can offer is a careful read of a shipping log, which is one of the more useful documents a product team produces in public: it records where the attention actually went.

Two rules we are holding ourselves to. We do not read a roadmap into silence — a month with no entry in some area is not evidence of anything. And where the board's own dating is ambiguous, we say so instead of picking whichever date reads better.

Safari and Edge are distribution work, not feature work

The 19 August release put Recall's extension on Safari (Mac) and Microsoft Edge, alongside the Chrome and Firefox builds it already had, and made it possible to add connections while editing a notebook inside the extension. It is tempting to file browser support under "features shipped". It is not a feature. It is distribution.

Porting an extension to Safari in particular is unglamorous engineering: a different packaging model, a different review process, and a long tail of behaviour differences that only surface once real users are on it. No team does that because it is interesting. A team does it because it believes the next tranche of growth sits behind a browser it cannot currently reach — Mac users who never left Safari, and the large population inside organisations where Edge is the browser IT installed and nobody changed.

Read alongside the rest of the same entry — a mobile share sheet on iOS and Android, drag and drop for PDFs, images, text files, Markdown files and Markdown zips — the shape is consistent. Most of 19 August was about widening the mouth of the funnel: more surfaces from which a thing can be got into the library. That is the kind of work that does not demo well and matters anyway.

Shipping OCR narrow, and naming the gap

The same release added what Recall calls images as cards: upload a JPEG, PNG, WebP or HEIC, and it reads the text in the image, describes the picture, and generates a title, summary and tags. One image at a time. Their own note says bulk images and OCR on social posts are "next up".

We do not think that deserves mockery. It is the correct decision, made in public. Image OCR into a knowledge base has a very wide quality range: a clean screenshot of a paragraph is easy; a photographed slide at an angle, a dense chart, or artwork with text baked into it are not. Shipping single-image first means the failure modes arrive one at a time, attached to a user who chose that image and can say what they expected. Shipping bulk first means they arrive fifty at a time, quietly, into a library the user then stops trusting.

Naming the next two steps in the release note is the part worth respecting. "Next up" is not a date and should not be read as one, but it tells a user which gap is known and which is not.

Chat suggestions, and the empty box problem

The third item worth reading closely: suggested questions on a card, a Suggested-questions panel in the desktop library, and starter questions for new users.

Be generous about this one. A blank chat box pointed at your own library is a genuinely hard interface problem in this category, and most of us have not addressed it. A user knows what is in their library the way they know what is in their own loft. What they do not know is what a retrieval system will do with a particular phrasing — and a bad first question returns a shrug, and after two shrugs people stop asking. Suggestions are an admission that the empty box was not working. Admitting that is worth more than another retrieval-quality claim.

Set it against the search release (titled 30 July 2026, dated 4 August on the board), which moved search out of the popup into the full-page library, searched reader content, notes and quizzes rather than card titles alone, and added a text-search versus AI-search toggle. Together with August's suggestions, that is a team treating "I cannot find the thing I saved" as an interface problem rather than a model problem. That is the right diagnosis.

One asymmetry in the same entry is worth flagging without snark: Add Content now accepts up to 10 links pasted at once, while bulk import of up to 1,000 bookmarks or 10,000 Markdown notes sits on the Plus plan, per Recall's pricing page on 4 September 2026. Ten is a paste-what-is-in-your-clipboard number, aimed at a different moment than migration, and that looks deliberate.

The write API, six weeks on

The 23 July release described the rebuilt social saves work — Instagram, TikTok, X, LinkedIn, Reddit, Facebook — as the "first step in the development of the write API". That is the single item on the board we watched most closely, because it changes what an assistant can do rather than what a person can click.

As of our read on 4 September 2026 there is no entry on the changelog board announcing a shipped write API, and Recall's own MCP documentation states: "The MCP server currently provides read-only access to your knowledge base. Write capabilities are coming soon." Their public-API URL redirects to that same MCP page, so there is no separate write-endpoint reference to check against.

That is an observation with a date attached, not a jab. A write API is a large piece of work — permissions, deduplication, idempotency, and the awkward question of what an assistant may put into someone's library without being asked. Six weeks from "first step" to not-yet-shipped is normal. We have written separately about what read-only means for anyone connecting Claude to Recall today.

Where BrainTube stands

The comparable parts, plainly. BrainTube runs one read-write MCP server covering the whole library, at https://mcp.brain-tube.com/mcp, and MCP access is included on every plan including Free — 44 tools as read from our own connect page on 4 September 2026. Where a paste box takes 10 URLs, our bulk_ingest tool is published as "Ingest up to 50 items in a single call with dedup and embedding", and an assistant calls it directly rather than a person pasting into a field. Screenshots are transcribed to text on ingest. We have a Chrome extension.

The parts we do not have. That extension is Chrome only — no Safari build, no Edge build — so the browser-coverage point in this post is one where Recall is straightforwardly ahead of us, and we are not going to claim otherwise. We have no native iOS or Android app; BrainTube installs as a PWA via add-to-home-screen, which is not the same thing as a share sheet wired into the operating system. Social-post capture is unverified on our side and we will not claim it. And we do no image or video generation at all.

How these figures were checked

Every Recall release fact above was read from their public changelog at https://feedback.recall.it/changelog on 4 September 2026, which returned HTTP 200 and rendered its entries in full; the newest entry on that board was 19 August 2026. The read-only quotation and the redirect from their public-API URL were read from https://docs.recall.it/developer/mcp on the same date. Plan contents and the bulk-import allowances came from https://www.recall.it/pricing, also read on 4 September 2026.

The search release is dated here the way the board dates it — titled 30 July 2026, carrying a board label of 4 August 2026 — because we could not establish which is authoritative. Nothing in this post rests on Recall's feature-request board: that page returned a JavaScript shell on our 4 September 2026 fetch with no vote counts rendered, so we left it out rather than reuse older numbers.

Claims about BrainTube describe functionality running in production on 4 September 2026, read from our own connect and pricing pages on that date. The next post in this series will cover whatever Recall ships between 20 August and the end of September 2026, read on the same terms. If you spot something here that has gone stale, tell us and we will correct it with the date attached.