Studio gains built-in site search and full-lifecycle automatic publishing, the Cloud CLI adds server-side command execution for environments, a new Resources page manages object storage, and AI spend limits now extend down to individual API tokens.
Cloud CLI v0.6 adds qc exec, which runs a command server-side in a one-off container for your environment. The run continues if your laptop sleeps or your connection drops. You can watch output as it runs, detach and re-attach later with qc exec status <run-id> --watch, and pick a container with --container. Runs are capped at one hour. The command uses the application-scoped Commands API, which the TypeScript SDK supports from v4.21.0. The documentation adds a qc exec reference and a remote commands guide.
A new Resources page in the dashboard shows your organisation's object storage and managed caches in one place. You can create object storage, attach it to and detach it from any application environment, and delete it. The credentials of attached storage take effect on the environment's next deploy. The environment receives them as environment variables, with an optional name prefix that you can set when you attach the storage. Managed caches attached to your applications are listed alongside, and templates that use object storage can provision it as part of a one-click template deploy.
Environment startup grace period and single-task setting in the API
You can now set a startup grace period for an environment, from the create and edit environment forms and through the API. It gives slow-starting applications time to become ready. During the grace period, failing health checks do not cause a newly started container to be replaced. The value is between 0 and 3600 seconds, and the default is 120. The single-task setting, previously available only in the dashboard, is now also accepted by the compose and create-environment APIs. This setting runs only one container at a time and turns off scaling for the environment.
You can now create a Payload CMS 3 application from a template. The template runs the Payload website starter with a managed PostgreSQL database, and keeps uploads on a persistent volume. Its secret key is generated when you create the application. Choose Payload CMS when you create an application from a template, or use the Deploy to Quant Cloud button in the template's README.
New QuantCode
Local OpenAI-compatible model providers
QuantCode can now use models running on your own machine. In the provider configuration, you can enable providers that have an OpenAI-compatible endpoint on a local loopback address (such as localhost), alongside Quant's hosted models. At startup, QuantCode notes that Quant does not audit requests to a local model. Endpoints on other hosts, including third-party services, are not supported.
New Studio
Built-in site search
Studio projects can now enable site search from the Search tab, in full page, modal or chat widget mode. Enabling search provisions an AI search index and asks the AI assistant to add the search widget to your site. If that does not finish, use Re-run setup. Published pages are indexed automatically when you publish, and deleted, renamed or unpublished entries are removed from the index. The Search tab also includes relevance boosts and excluded paths, a Purge index action, an in-place re-index with live progress, and a search preview.
New Studio
Per-project AI instructions
A new AI Instructions tab in Studio settings lets you record standing rules for the AI assistant — for example, which styles are global and which page copy must not change. Instructions are added to the assistant's context on every turn, and an AGENTS.md file at the repository root is picked up automatically as well.
Scoped API tokens, introduced in March, gain a content:purge scope for purging cached content. Deploy and CI tokens no longer need projects:write just to purge. That scope also allows editing projects, domains and headers. Tokens created with the Content Management preset now include the new scope, and tokens that purge with projects:write today keep working.
Use the content:purge scope instead of projects:write for deploy and CI tokens that only need to purge.
Improved AI platform
Per-token spend limits
Following June's per-interface caps and per-user overrides, AI governance budgets can now be set per API token, with daily and monthly limits. The governance dashboard shows token names on override rows and suggests your organisation's tokens as you add them, and the new fields are exposed through the API and SDKs. New documentation pages cover AI governance and filter policies.
Each workflow now has an Engine setting with three options: Auto, Durable and Inline. Auto is the recommended default: it runs workflows that wait on steps with the durable engine, and runs all others inline. Inline runs are faster, but they do not wait for asynchronous steps, so later steps do not get the results of those steps. Tool steps that await their result gain two settings. One sets how often to check the job (minimum 15 seconds, default 45), and the other sets how long to wait before the step times out (default 60 minutes, maximum 24 hours). Orchestration steps can now run a tool or another workflow over their input items, as well as an agent.
Improved AI platform
OpenAI compatibility guide
A new OpenAI compatibility page in the documentation shows how to use the /oai/v1 endpoints introduced in June with OpenAI SDKs and tools. It covers the base URL, authentication, the supported endpoints, streaming, tool calling and model IDs, with quick-start examples in Python, Node and curl. It also lists the differences from the OpenAI API, including the features that the endpoints do not support.
The documentation now has a full command reference, with arguments, options and examples, for both command-line tools. The Content CLI (quant) reference covers every command. The single Cloud CLI (qc) commands page added in April is now split into one page for each command group. New overview pages explain which CLI to use for which task. The Content CLI pages have moved from /cli/ to /content-cli/, and the old addresses redirect to the new ones.
Managed cache metrics, costs, and template support
You can now also add June's managed cache, which is Redis-compatible (Valkey), when you create an application from a template, a pre-built application or a GitHub repository. The same storage cap setting applies. After a cache exists, the application's Environments page shows it as enabled, with a reminder to redeploy your environments until the cache is connected. A new Cache tab on the environment metrics page charts memory use, hit rate, processing units and connections. Cache costs appear as line items on the cloud usage page, at a flat monthly rate based on the configured storage cap. Templates can recommend a cache so that it is provisioned during template deploys. A new documentation section covers the cache.
Improved Quant Cloud
PHP performance settings in the PHP template image
Applications built on the Quant PHP base image, including the Drupal and WordPress templates, can now tune PHP's opcode cache with PHP_OPCACHE_* environment variables. The defaults are unchanged. Files that are generated at run time are excluded from the cache, so you can turn off file timestamp checks safely. Static files such as images, CSS and JavaScript now send 7-day cache headers by default. The CDN then serves them instead of your application storage. Set QUANT_STATIC_CACHE=0 to turn the headers off.
Rebuild your application to apply the new PHP and static file cache settings.
Improved QuantCode
Intel Mac support
QuantCode now ships Intel (x86_64) macOS builds of both the CLI and the desktop app alongside the existing Apple Silicon builds, including per-architecture desktop installers and Homebrew support.
Improved QuantCode
Autonomous jobs run longer and preserve their work
Autonomous jobs can now run for up to 90 minutes, up from 60. If a job reaches its runtime or spend ceiling, its work is committed, pushed, and opened as a draft pull request automatically, so work in progress is preserved rather than lost.
Improved QuantCode
Automatic model fallback
If a Quant-hosted model declines a request, QuantCode now continues on the suggested fallback model automatically, with a note in the session showing which model answered. The fallback applies for the remainder of the session, and the model selector reflects the change.
Improved SDKs
TypeScript client v4.21.0 and Go client v4.20.0
Both clients add the OpenAI-compatible endpoints introduced in June (chat completions, embeddings, and model listing), per-token AI spend limits, managed cache details on applications, and the environment single-task and startup grace period fields. The TypeScript client also moves the Commands API to application-scoped paths, so its methods now take an application ID, and adds optional container targeting, as used by the new qc exec command.
Pass an application ID to the Commands API methods when you upgrade the TypeScript client to v4.21.0.
Improved Studio
Automatic publishing across the content lifecycle
Auto publish on save is now an explicit opt-in setting for Quant Builder mode projects. It covers the full content lifecycle on your production branch. Saving an entry publishes its page. Deleting an entry or switching it to draft asks for confirmation, then removes its page from the live site. Renaming an entry removes its old page, and an entry that republishes at a new URL cleans up its old one. With auto publish on, media uploads from the media browser, the AI assistant or the API now sync to the live site immediately. Changed media is purged from the edge cache, and deleted media is removed from the live site. A publish-status chip in the entry editor shows each entry's live state, with retry and view-live actions. Listing pages, feeds and the sitemap update on a full site publish, which you can now run in one click from the entry editor. A banner on the Deployments page warns when code or content has changed since the last full publish.
Turn on Auto publish on save in Studio settings to use it, because it is now off for all projects, including existing ones.
Improved Studio
Review and merge workflow
A Review button on the Branches screen opens a pull request for any working branch, or reuses an open one, and takes you straight to the review. On the review page, Ask AI to fix posts the comment you typed and starts a focused assistant conversation to address it, and the reply appears in the comment thread when the assistant finishes. In Quant Builder mode, merging a pull request no longer starts a separate GitHub Actions build; the review page prompts you to Build & Publish the merged changes instead. The file browser and Deployments toolbars also gain their own branch selector.
Improved Studio
Published output closer to a standard Astro build
Publishing in the browser with Quant Builder mode now matches a standard Astro 6.4 build more closely, checked against Astro's own output with a parity test suite. Component styles are now scoped with Astro's default attribute strategy, so elements no longer carry astro-* classes and scoped selectors are more specific. Published sites include a generated sitemap when your Astro config sets a site URL, but custom sitemap filter and serialisation options are not applied. Images are emitted as hashed asset files instead of being embedded in pages. Inline styles that you write, font and favicon links, and custom Tailwind plugins are preserved. Pagination and dynamic routes render correctly, build.format and trailing-slash settings are respected, and editor markers are stripped from published markup.
Update any CSS or scripts that target astro-* classes before you republish.
Opening a large content collection is now faster. The list loads more entries as you scroll, with a loading indicator. Opening an entry in the editor is faster too. Edits outside Studio now appear in the list within 30 seconds, rather than up to 5 minutes. Collections with a date field now sort newest-first by default, and a sort dropdown offers newest, oldest, A–Z, and Z–A ordering, remembered per collection.
Improved Studio
Claude Opus 5 and Sonnet 5 in the AI assistant
Claude Opus 5 and Claude Sonnet 5 are now available in the assistant's model selector, and Opus 5 is the new default. Claude Opus 4.8 and Opus 4.7 replace Opus 4.6 in the selector. The model you select is now used for the whole response, and a new Studio default option is available. If the selected model declines a request, the assistant resubmits your message once on a fallback model, in the Studio assistant and in the edge function editor's AI assistant.
Improved Studio
AI assistant attachments and site awareness
The assistant can now see images: media library files are shown to it as actual images rather than filenames. It can save an image you attach in chat straight to your media library. Image attachments are carried through multi-step work instead of being dropped after the first step. Large images are downscaled in the browser before sending, with a clear error if a file is still too large. A computed summary of your site's architecture — shared components, layouts, global styles, and collection templates — is added to the assistant's context, so it can scope local versus global edits. The assistant now asks you to confirm before it edits a shared file. If the assistant's session has expired, it now resends your message in a fresh session instead of showing an error.
Improved Terraform
Per-token AI spend limits
Following June's per-interface caps and per-user overrides, the quant_ai_governance resource now supports per-token spend limits. Flat daily and monthly caps apply to every API token without its own override (per_token_daily_budget_cents, per_token_monthly_budget_cents). Named per-token overrides, with optional unlimited exemptions, are set with token_overrides. The provider has also been updated to use version v4.20.0 of the API client.
Improved WAF
Configuration guide
A new guide walks you through configuring the WAF on a proxy rule: start in report mode, review WAF logs, tune false positives with allowed rule IDs and IPs, then switch to block mode. It covers paranoia levels 1 to 4, block lists for bad bots, referers, IPs and AI crawlers, and IP and header-based rate limiting. It includes Slack notifications and recommended settings for WordPress, Drupal, API and static sites.
If the project you last had selected was deleted, every dashboard page could fail to load. Pages now load normally, and removing a Studio environment you are working in switches you back to its parent project.
Fixed GitHub Actions
Custom AI tool deployment
The agent tools action, quantcdn/quant-cloud-agent-tools-action, failed to register custom tools: every deployment was rejected with a validation error. Deployments now succeed, and existing tools are updated in place. The platform deploys the function and registers the tool in one step, so you no longer need a separate deploy step. The action no longer uses the preview_domain and quant_project inputs.
Add each tool's edge function source as fn.js next to its tool.json, and remove the separate edge function deploy step.
Fixed Projects
Container selection when creating a CDN project
When you created a CDN project from a Quant Cloud environment, the container and port dropdowns could show no options. They now populate correctly, and each dropdown shows a loading indicator while its options load, instead of showing nothing.
Fixed Studio
Long AI turns complete reliably
Long-running assistant responses could fail before completing, and heavy multi-file turns could end with a blank reply. Long turns now complete reliably, with a progress indicator while the assistant works and a Retry button that replays a failed turn in the same conversation. The assistant can now read many files in a single step, and a turn always ends with a reply. If it reaches its step limit, it summarises the changes made so far and you can ask it to continue.
Fixed Studio
Builder stability on large sites
Building a large site in the browser could crash the tab part-way through. Large projects now build reliably, and a single page that fails to render no longer stops the whole build. React components now also render correctly in the preview on first load.
Fixed Studio
Re-imported environments publish
Deleting a Studio environment and then re-importing it could leave the new environment unable to publish, and its preview URLs could fail to load. Re-import now restores the original environment, so it publishes and its preview URLs resolve correctly.
Fixed Studio
Entry editing and production branch fixes
Saving an .mdx or .yaml entry no longer creates a duplicate .md copy, and you can now delete locale-nested entries from the collection browser. Publishing, purging and media sync now work when your Studio production branch is not the repository's default branch.
Fixed Studio
AI assistant pull requests target your production branch
Pull requests opened by the AI assistant always targeted main, even when your production branch has a different name. They now target the production branch in Studio settings, and the assistant asks you to switch to a working branch if you are on the production branch.
Fixed Studio
Publishing resilience
A brief network interruption while publishing could cause a whole batch of pages to be reported as failed. One interrupted upload no longer fails the rest of the batch.
Fixed Studio
Code editor TSX and JSX support
.tsx and .jsx files showed false syntax errors on every JSX construct and flagged imports as unresolvable. The code editor now parses JSX correctly, and each open file keeps its own undo history.