All months

Changelog · 2026-09

September 2026

Studio gains a component editor, an Insert palette, undo and redo, and the CivicTheme template; Quant Cloud environments can sleep when idle; rules can serve custom error pages by path; and edge functions gain expiring KV keys, queries and counters.

52 updates across 12 products shipped this month.

New 9

Back to top
New KV stores

Expiring keys, queries and counters in edge functions

Edge functions can now make a key expire: pass ttl in seconds to kvStore.set(), from 60 seconds to one year. An expired key reads as null, writing the key again without ttl removes its expiry, and the API returns the expiry as expires_at. New kvStore.list() and kvStore.search() methods page through a store with a cursor, and search() filters by key prefix, a * key pattern or exact matches on top-level JSON fields. kvStore.increment() adds to a numeric key atomically, so concurrent requests do not lose each other's updates, and it accepts a ttl for counters that should expire.

Related documentation: Key Value Stores

New KV stores

Clear a store in bulk

A Clear store action on the KV stores page deletes keys in bulk. You can delete every key or only the keys with a given prefix. You can also delete only the keys last updated more than 1 hour, 24 hours, 7 days or 30 days ago, or before a date you choose. You type the store name to confirm, and large clears continue in the background. The API supports the same bulk delete with DELETE /api/v2/organizations/{org}/projects/{project}/kv/{store_id}/items and its prefix and older_than parameters. Deleting a store that still holds keys now tells you to clear it first, and the API accepts force=true to delete a store together with its data.

New Quant Cloud

Idle sleep

Environments can now stop their containers after a period with no requests and start again on the next visit. Turn on idle sleep from the environment's Idle sleep control or through the API, and choose an idle period from 10 minutes to 24 hours (30 minutes by default). The first visitor after a sleep sees a loading page for one to three minutes while the environment starts. Only compute sleeps: databases, storage and caches keep running. Environment lists show when an environment is sleeping or waking. Idle sleep suits editorial environments whose public site is served from the edge. If an environment cannot use idle sleep, the API returns the reason.

Related documentation: Idle Sleep

New Quant Cloud

Resources API

The object storage and managed caches on the Resources page can now be managed through the v3 API, so a CI pipeline can provision an environment's storage and cache with the rest of its configuration. New endpoints list, create, view and delete organisation resources, and attach them to or detach them from an application environment, with the same permissions as the dashboard. The responses are typed in the API specification, so generated SDK clients return resource and attachment models.

New Rules

Custom error pages by path

A new Custom error page rule action serves a published page that you choose when a matching request ends in an error. For example, a rule can serve /fr/_quant404 for missing pages under /fr/*, so visitors to a multilingual site see an error page in their language. The response keeps its original status code. The action applies to not-found pages and to errors from proxied origins. Requests that match no error page rule, or whose chosen page is not published, still get the project's usual error page. You can set it up in the rule editor or through the new error page endpoints in the Rules API.

Related documentation: Page Rules: Error page

New Search

Custom sort order

A Sort Order card on the search configuration page lets you add ranking rules. Each rule is a field and an ascending or descending direction, and it applies after text relevance, for example to list newer pages first among results that are equally relevant. The rules apply to your project's search results, including dashboard-hosted search pages. On QuantGov, dashboard-hosted search pages apply the first two rules.

Related documentation: Enable Quant Search: Sort order

New Studio

Insert components and reorder blocks

An Insert button in the entry editor opens a searchable palette of your site's components, grouped by category. The component you choose opens in the component editor with its default settings filled in, and Insert adds it after the selected block. A component appears in the palette when it is listed under _components in quant.studio.json or tagged @studio in its file. An empty palette explains these two options. To add a component from the file tree, choose Make insertable in its file's menu. To reorder body blocks, drag them in the preview. While you drag, the preview shows where the block will land. Press Escape, or drop the block where it started, to restore the order.

New Studio

CivicTheme template and Sass support

The CivicTheme design system is now available as a Studio template from Start from Template, and that page now opens at once and lists every Studio-ready template. Studio now compiles Sass: .scss files appear in the file tree, .scss imports and <style lang="scss"> blocks compile in the preview and in published builds, and editing a partial refreshes every page that uses it. The preview now has a white background, so light-themed sites are readable. If a template has its own search page, Studio keeps it when you enable site search.

New Studio

Undo and redo in the visual editor

The entry editor now has Undo and Redo buttons and the standard keyboard shortcuts for changes made in the preview. They cover inserts, component edits, moves, formatting, image and link edits, inline text edits, AI edits and Revert. Studio keeps up to 100 steps per entry and announces each step to screen readers. Undo and Redo are off in Code and Split view, which keep their own undo history.

Improved 21

Back to top
Improved Crawler

Custom user agent with browser mode

You can now set a custom user agent on crawls that use browser mode, from the dashboard, the API, Terraform or Pulumi. Before, the dashboard field was turned off whenever browser mode was on. Browser-mode crawls without a custom value now send the crawler's default user agent, and clearing the field returns a crawler to that default.

Improved Crawler

Run tracking through the API and Terraform

Run tracking, which provides the per-run details added in August, can now be turned on when you create or edit a crawler through the API. The new tracking field is also returned when you read a crawler. The Terraform quant_crawler resource has a matching tracking attribute, so you can turn tracking on for many crawlers at once from Terraform or Pulumi.

Improved Crawler

Stale content review with exclude patterns

Stale content review, introduced in August, now removes pages that match one of the crawl's exclude patterns without re-checking them against your origin. These pages also no longer count towards the limits that hold a large removal for your approval. Single-URL and URL-list crawls now skip the review. Re-checks now work with origins that reject HEAD requests or stop responding. If the origin rejects the HEAD request, the re-check tries once more with a GET request. A crawl that finished is no longer reported as failed while its review runs. Pages held back from removal stay tracked, so a later crawl can remove them.

Improved Edge functions

Time zones for scheduled functions

Scheduled edge functions, introduced in November 2024, can now run their cron expression in a time zone you choose, such as Australia/Sydney, instead of always in UTC. A 02:30 schedule then runs at 02:30 local time. The schedule form uses your browser's time zone by default, the schedule list shows each schedule's time zone and calculates the next run in it, and the API accepts an IANA time zone name. Schedules without a time zone still run in UTC. New schedules run the function once for each scheduled time, without retries, and the cron builder now shows the saved hour correctly when you edit a schedule.

Improved Quant Cloud

Resources in every create flow

Every way to create an application, from a template, a GitHub repository, a public image or the manual form, now has a Resources step for object storage and a managed cache. In that step you can create a new organisation resource, attach one your organisation already has, or skip it. When a template recommends a resource and your organisation already has one, that resource is selected for you. Resources chosen at creation are applied to the new environment without a manual redeploy, and the application's Environments page lists attached resources with attach and detach controls. The dashboard sidebar also groups Applications and Resources under a new Cloud menu, and New Application has now moved to the Create menu group.

Improved Quant Cloud

Cache purge controls and admin access

The Resources page adds three cache controls. Purge keys, on each cache attachment, deletes only that environment's keys. Flush entire cache asks you to confirm before it flushes the whole cache. Admin credentials shows the credential for the whole cache. The same operations are available through the API. Environments attached to an organisation cache in September or later cannot flush the whole cache themselves. Frameworks that clear the cache with a full flush, such as Laravel's cache:clear, therefore get a permission error.

Replace full cache flushes, such as Laravel's cache:clear, with prefix-based or tag-based clearing or Purge keys, or attach the environment with the Admin access level.

Related documentation: Organisation Caches: Administrative access and purging

Improved Quant Cloud

Deployment failures shown while retrying

When new tasks for an environment cannot start, for example because the image tag does not exist, the environment page now shows the failure and its reason while the platform is still retrying. Before, the failure appeared only minutes later, when the deployment was marked failed. The Logs tab stays empty until a container starts, so this is the first sign of the problem. A task reclaimed from spot capacity now shows as a warning that needs no action from you, and the image-not-found hint explains that the prefix of the image tag must match the container name in the Docker Compose definition for your application.

Improved QuantCode

Autonomous jobs start faster and recover from stream errors

Autonomous jobs now default to Claude Opus 5.5 and no longer build a code index at start-up, which saves about three minutes per job. A model stream that stalls or fails part-way is retried instead of failing the job. A job that is stopped still sends its final report, and subagent commands run in the cloned repository.

Improved QuantCode

Separate attachments model and reasoning off

You can now set a separate attachments model that reads and summarises images, PDFs and audio when your main model cannot read them. Self-hosted Qwen3.8 and DeepSeek V4 models also gain a reasoning-off option alongside their effort levels, which you can choose turn by turn. It gives much shorter responses on simple turns. It is not the default, because hard turns do better with reasoning on. Reasoning effort levels and the low-effort default for Qwen3.8 now also apply to a self-hosted model with a custom name when its configuration sets the model family.

Improved QuantCode

Repeated tool calls stop the turn on self-hosted models

On self-hosted models, QuantCode now stops a turn that makes the same tool call with the same input three times in a row and gets the same result each time. A model stuck in a loop no longer runs until you interrupt it. Calls whose results change, such as rereading a file after an edit, do not count. Quant-hosted models are not affected, so polling a build or deployment with unchanged output never ends the turn.

Improved SDKs

Go client v4.24.0

The Go client moves from v4.20.0 to v4.24.0. It adds the organisation resources endpoints, including cache credentials and purging, environment idle sleep, custom error page rules, bulk deletion of KV keys, and the crawler tracking field.

Improved Studio

Component editor saves to the page

Clicking a component in the preview now opens a component editor that writes your changes into the page source. It changes only the settings you edited. The props panel introduced in April changed only the preview. The editor shows a live preview of the component with your site's styles at mobile, tablet and desktop widths. Its inputs follow the component's prop types: text, Markdown, images, page links, icons, choices and lists. A breadcrumb, a Contents list and a Components menu take you to parent and nested components. The image input accepts a typed address as well as a picked file. Settings that are set in code are read-only, with a link to Code view.

Improved Studio

Automatic publishing covers more saves

Automatic publishing, introduced in July, now covers more changes. It publishes saves from the Files view, which now shows the same publish-status chip, with Retry. It publishes a renamed or moved page at its new address, and it publishes the pages that the AI assistant changes in a turn. When the AI assistant, the API or the file tree deletes an entry, automatic publishing removes the live page. Saves made while a publish is waiting share one site build, and an AI turn that changes several pages publishes them in one build. Entries in collections that have no pages skip the build. When publishing is off for the project or the branch, the button reads Save instead of Save & Publish, and the confirmation message says that publishing is turned off.

Improved Studio

Publishes send only what changed

Skipping unchanged files on publish, introduced in March, now uses publish records kept with your project instead of in your browser. Pages and media that are already live and unchanged are skipped whichever browser or device publishes. Media uploads now run ten files at a time. A save syncs only the media that its pages use, including images named in component settings and stylesheets, so a file that cannot be uploaded no longer makes unrelated saves fail. A page is published only after its media is on the CDN. A save made while another media sync runs waits for that sync instead of failing, and a failed sync names the files it could not upload. The first publish of each project and branch after this change sends every file once.

Improved Studio

Social images, feeds and redirects published

Studio publishing now produces more of what a local astro build produces. The redirects in your Astro config, including maps imported from another file, are published as redirect pages. Like a static astro build, these are HTML pages that redirect in the browser, not HTTP redirects. Endpoint routes, such as an rss.xml.ts feed, are run and published. Some endpoints, such as those that generate social (Open Graph) images, need packages that a browser cannot run. These are now built on the server with your project's own dependencies after each publish. Only the files that changed are uploaded and refreshed at the edge, and the Deployments page shows how many files were published and how long it took.

Improved Studio

Stale pages removed after a complete build

Update now, the one-click full publish, also unpublishes pages that your build no longer produces, such as a page whose source file was deleted outside Studio, and removes them from site search. It removes pages only when every page built. If some pages fail, it publishes what it built, keeps the other live pages, and names up to three failed pages with the build's own reason. Routes the browser cannot build, such as social image endpoints, no longer mark a run as failed, and their pages are kept. The home page is never removed.

Improved Studio

AI edits to the open entry wait for review

When the AI assistant changes the entry you have open, nothing is written until you accept the change. The proposal appears as a diff against your current text: Accept puts it in the editor for you to save, and Reject leaves the editor and the repository unchanged. The assistant cannot delete the entry under review, and a proposal is refused if you have unsaved edits or if the reply was cut off. Edits to other entries and other files are still written directly.

Improved Studio

Saves change only what you edited

Saving an entry now rewrites only the frontmatter fields you changed, and keeps the comments, quoting and formatting of every other field. A save that changes nothing makes no commit and shows "No changes to save". A failed save is reported and keeps your changes. When an unsaved draft replaces the saved version as you open an entry, a notice says so and offers Keep editing and Discard.

Improved Studio

Image fields

The Fields panel now offers the image picker for any field whose name shows that it holds an image, such as bannerImage, logoLight or ogImage, not only for a fixed list of names. Browse opens at the folder of the current image, and absolute image URLs show a preview. An alt text box appears only where your schema keeps alt text or the entry already has alt text, and a missing required alt is flagged before you save. Image fields you did not edit are saved as they were, except that a plain path in a field whose schema expects an image object is saved in that object form. If the media browser cannot list a folder, it now shows an error instead of an empty folder.

Improved Studio

Creating translations

Create translation now refuses when the translation already exists, instead of overwriting it, and the locale bar updates as you move between entries. A translated copy's path or permalink gets its locale prefix, so the copy no longer replaces the original page. From March until this change, creating a translation also wrote a locale page route into your site. Studio no longer does this. If no existing route builds the translated pages, Studio still creates the translation, and a note says that your site may need a route.

Improved Studio

Claude Opus 5.5 in the AI assistant

Claude Opus 5.5 is now available in the AI assistant's model selector and is the new default, ahead of Opus 5.

Fixed 22

Back to top
Fixed AI platform

Long streamed responses

Streamed chat responses from the AI API stopped after about five minutes, so a model that reasoned for several minutes before its first token returned no content. Streams now continue for up to 15 minutes.

Fixed AI platform

Custom tool starter template

Custom tools written from the default template in the AI tool editor failed every time they ran. The starter template now uses the edge function format and runs as written.

Fixed Cloud CLI

Clear error for qc ssh on Apple silicon

On Apple silicon Macs without Rosetta, qc ssh failed with "spawn Unknown system error -86" when a required tool was installed as an Intel build. Cloud CLI v0.6.2 now explains the mismatch and names the native (arm64) build to install.

Fixed Crawler

Crawl delay and missing files

Crawls that do not use browser mode ignored the configured delay between page requests. The delay now sets the minimum gap between page requests across all workers, so a delay of 4 seconds allows about 15 page requests a minute. Crawls with a delay can take longer than before. Static copies now include the AVIF versions of images offered in <picture> elements, because .avif is now in the default allowed file extensions. A single URL, start URL or sitemap entry that points to a file, such as a PDF, is now captured.

Fixed Edge functions

Cookies and empty responses

When an edge function set more than one cookie in its response, only one reached the browser. Every Set-Cookie header is now kept, including those from inline functions in proxied pages. A content filter on a function response with no body, such as a redirect, a 204 or new Response(null), could fail the request; such responses now pass through the filter unchanged.

Fixed KV stores

New stores say when they are ready

For about 30 seconds after you created a KV store, listing or adding items showed a generic failure. The dashboard and the API now say that the store is not ready yet, and the Terraform quant_kv_store resource waits until the store is ready before reporting it as created.

Fixed Quant Cloud

CPU metrics chart

The CPU chart on the environment metrics page could show peaks near 1000%. It now shows CPU as a percentage of the environment's allocated CPU, and the minimum, average and maximum figures cover the whole selected period instead of only the latest data point.

Fixed Quant Cloud

PostgreSQL ownership after a database sync

A PostgreSQL database copied to another environment by a database sync, including when you clone an environment, could keep tables and other objects that the target environment's database user did not own. Migrations in that environment could then fail. Copied objects now belong to the target environment's database user, so migrations no longer fail for this reason.

Fixed QuantCode

Cloud tool results reach the model

Results from cloud tools, including web search, web browsing, task management, skills and your organisation's server-side custom tools, did not reach the model, which reported the tool as interrupted. Results and tool errors now reach the model.

Fixed QuantCode

Compacting long sessions

A long session could fail with "Session too large to compact" while the conversation itself still fitted the model, which left the session unusable. Compaction now succeeds on these sessions.

Fixed QuantCode

Recent sessions in the session list

New sessions could disappear from the QuantCode app's session list after a few hours, which looked as though they had been deleted. The list now shows your most recent sessions.

Fixed Studio

Publishing to the live site

Images, fonts and other media could disappear from the live site after a publish and stay offline after later publishes. Media now stays published. Saving an index entry, such as the entry a site shows as its home page, showed No page and left the live page unchanged. Index entries now publish to their folder address (/ or /docs/) when that page shows the entry. Unpublishing or deleting an index entry never removes the home page or a section page. Each Studio environment is a child project of the Studio project, which is its parent project. When an environment was mapped to a project's production or default branch, publishes from that branch went to the parent project instead. Each branch now publishes to the environment mapped to it.

Fixed Studio

Automatically published pages use saved content

Pages published automatically on save could differ from the same pages in a full site build. They are now made by the same build as Update now and use only saved content, for the live site and for the search index. A publish that the CDN does not accept is now reported as failed instead of Published. A page whose content does not compile is not published, and Studio shows the reason. An entry whose frontmatter Studio cannot read in full is not published until you fix the line that the message names.

Fixed Studio

Edits are kept and stay with their entry

Edits in the Fields panel could be lost when you reloaded or left the page, and a later visual edit or save could overwrite text typed in Code view. If you switched entries or locales, used browser Back or opened another file while an edit, a save or an AI reply was in progress, the change could go to a different entry. Fields panel edits are now kept as the entry's draft, Studio asks before you leave with unsaved changes, and Code view edits are copied into the entry. Edits, saves, reverts and AI proposals now apply only to the entry they were made for, and the AI assistant keeps one conversation for each entry. A draft is restored only over the version it was based on, and a draft that cannot be read is kept, so you can fix it in Code view or discard it with Revert.

Fixed Studio

Inline editing in the preview

Editing text in the preview could change an earlier copy of the same words, lose text typed after Enter, break bold, link and code formatting, or write non-breaking spaces into your content. Bold, Link and Wrap could also apply to the first copy of the selected text, not the one you selected. Edits now apply only to the text you edited and keep the formatting around it, Enter finishes an edit, and Shift+Enter adds a line break. If Studio cannot find the exact source text, it refuses the edit with a message instead of saving it in the wrong place. Clicking text inside a component opens the component editor, and inline edits to header fields such as banner titles now save.

Fixed Studio

Multilingual projects

On projects with more than one locale, Studio could list only some entries of a collection and miss entries in subdirectories. It could also save edits to a new locale copy instead of the original file, and create new entries in the wrong folder. Studio now reads the locale settings from your Astro config, lists and saves each entry from its own file, and tells you when the selected locale hides entries. Locale folders match their locale in any letter case, such as pt-br for pt-BR, and locale badges follow the project's locale order. Deleting a translation no longer removes the default-language page, and AI assistant turns on multilingual sites no longer stop with "Stream ended unexpectedly".

Fixed Studio

Entries keep their schema and format

New entries in collections whose schemas use list defaults, shared schemas or nested objects could get values that the schema rejects, such as a list field saved as text. The whole site then failed to build. Fields from shared schemas or nested objects could also be missing or in the wrong place in the Fields panel. New entries now get values of the correct type and the file extension that the collection loads (.md or .mdx). A save that would leave out a key the schema requires is refused, with the key named. AI assistant edits to an .mdx entry now update the entry's own file instead of creating an .md file that the site does not load. JSON and YAML data entries now load and save in their own format without added frontmatter and have no Fields panel, and their fields are now indexed for site search.

Fixed Studio

Preview start-up

The preview could stay blank, show a render error for a file that is in the branch, or report Ready when it had failed to start. A preview that fails to start now retries once automatically and then shows the reason with a Try again button, and the status bar reports Ready only when a page is on screen. Editing a content entry now refreshes the home page preview.

Fixed Studio

Collection lists show every entry

In a long collection, the grid could stop loading entries part-way, Load more could keep spinning, and collections in very large repositories could be listed only in part. Loading now continues until every entry shows, a stalled request times out so that you can try again, and keyboard focus moves to the first newly loaded entry.

Fixed Studio

Published site fixes

Visitors to a missing page on a Studio site saw the generic error page instead of the site's own 404 page. Studio now publishes your rendered 404 page as the site's error page on every publish, with nothing to configure. On published sites, tables of contents built from a page's headings now render, and quoted frontmatter values with escapes or line breaks now display correctly. Files in hidden folders under public/, such as .well-known, are now published, and tsconfig.json path aliases now resolve. Previous and Next pagination links showed as disabled on every page and now link to the correct pages. On sites that use Astro view transitions (<ClientRouter />), the browser's Back and Forward buttons changed the address but left the previous page on screen. They now show the page that matches the address. Published sites got a sitemap whenever the Astro config set a site URL. They now get one only when the config uses the @astrojs/sitemap integration, and the sitemap matches what the integration writes, including its filter and serialisation options.

Add the @astrojs/sitemap integration to your Astro config if your Studio site needs a sitemap.

Fixed Studio

Editor fixes

The Deployments page now loads on a branch with no mapped environment, so you can publish from that branch. The browser tab title now changes when you switch entries, and saving a form keeps the form settings that the form editor does not show. The AI panel no longer squeezes the editor on screens 1024 pixels wide or narrower. The body context menu in the preview now works from the keyboard and stays inside the window, and the Save & Publish menu button has an accessible name. After a pull request merges, the review page no longer offers to delete the branch, so a long-lived branch is not removed by accident.

Fixed Terraform

Provider fixes

quant_ai_custom_tool resources can be created and updated again, after every create and update failed with a validation error. Applies that change several rules on one project no longer fail part-way. Projects, domains, custom headers, crawlers and crawler schedules deleted outside Terraform are now removed from state instead of failing every plan. Changing the name of a quant_kv_store or the host of a quant_domain now replaces the resource; a domain change previously reported success without changing the live record. Crawlers that set the assets block apply without a value conversion error, and crawlers with an empty start URL, include or allowed domains list now refresh all settings. Under Pulumi, quant_crawler no longer shows a permanent diff and quant_application can be created.