Datasette 1.0a43 is out, and it moves Simon Willison’s open-source data exploration tool closer to its long-awaited stable release. Willison reports that the 1.0 milestone is now 53% complete. He credits a lot of that progress to Claude Fable 5.2, which he’s been “leaning hard into” to work through the backlog.
It’s an alpha release, but it’s a big one. It brings a major architectural feature, two security fixes and a long list of smaller improvements. Here’s what’s in it.
Streaming support for output renderers
This is the biggest 1.0 item in the release. Datasette’s built-in CSV and TSV exports now stream their output instead of building the whole response first. Plugins can use the same approach through a new Response.stream() method, which lets them register as able to stream every row in a table in one go.
Why it matters: Datasette is often used to publish large datasets. With streaming, a plugin that exports to a custom format can handle huge tables without running into memory or timeout limits. Locking this API in before 1.0 gives plugin authors a stable target to build on.
A security fix with a GitHub advisory
The release includes two security fixes. Willison judged one of them serious enough to publish a GitHub advisory. Private stored queries could be exposed, revealing the title and SQL of the query to people who shouldn’t see them. The flaw didn’t let anyone actually run those queries.
It’s limited in scope, but SQL can reveal table structures, business logic or sensitive filter values. If you run a Datasette instance with private stored queries, upgrade.
A row page JSON format built for 1.0
The JSON representation of individual row pages “finally has a design suitable for 1.0,” according to Willison. For anyone building on Datasette’s API, this is the kind of quiet change that matters most. A stable JSON shape means integrations won’t break once 1.0 ships.
A batch of UI improvements
There are dozens of smaller upgrades. Some of the more useful ones:
<datasette-copy-button>: a new Web Component that adds copy-schema buttons in several places- Pretty-printed JSON: JSON stored in columns can now be formatted for easier reading on the row page
DatasetteModal.confirm(): a new JavaScript utility for confirmation dialogs- Dozens of bug fixes, including contributions from Zihan Dai, Peter Olkhovets, Ilya Khalafi and Ankita Advitot
View transitions make it feel like an SPA
The feature Willison says he’s most excited about is “a pretty tiny one.” Datasette now uses CSS cross-document view transitions. When you click from page to page, elements like facet results and data tables animate smoothly from their old position to their new one instead of jumping.
It only takes a few lines of CSS. A @view-transition { navigation: auto; } rule turns on the effect, wrapped in a prefers-reduced-motion media query so people who’ve asked their system for less motion don’t see it. Facet results and the main table each get a view-transition-name, which lets the browser morph them across page loads.
The result is that Datasette, “still staunchly a multi-page application,” now feels more like a single-page app. I think this is the most interesting idea in the release. For years, teams reached for heavy JavaScript frameworks partly to get smooth navigation. Native browser view transitions let traditional server-rendered apps get much of that polish almost for free.
One caveat: it currently works in Safari and Chrome. Willison expects Firefox support to arrive in Firefox 162, due December 8, 2026, since the feature is no longer behind a flag in Firefox Nightly.
The AI angle
Willison’s note about Claude Fable 5.2 is easy to skim past, but it’s worth noticing. This is an experienced maintainer using a frontier model to make steady progress on a mature open-source project: streaming APIs, security fixes and UI polish. It isn’t a flashy demo. It’s AI as a productivity multiplier on real, long-lived software, and that’s where these tools are proving their value.
What’s next
At 53% of the milestone, Datasette 1.0 still has work ahead, but each alpha locks in more of the plugin and JSON APIs. If you build plugins, start testing against Response.stream() now. The full release notes and the advisory details are in Simon Willison’s original post.