KV stores gain format 2 with indexes, transactions and append-only ledgers, edge rules can be managed in code, and the month adds an AI audit log, a Slack bot messaging API, a DNS API, Studio publishing roles and content API token rotation for static projects.
The open-source Quant agent skills teach QuantCode, Claude Code, Codex and other AI coding agents to work with QuantCDN and QuantGov. With the skills installed, an agent uses the right commands, endpoints and settings for your platform instead of guessing. The 17 skills cover static sites, Quant Cloud applications, starter templates, rules, edge functions and KV stores, CI pipelines, custom domains, and caching and purging. They also cover Terraform and Pulumi, the Drupal and WordPress integrations, Search, AI services, Studio and the Slack bot messaging API. A security best practice skill sets out the order in which to secure a project. It covers origin protection for Quant Cloud applications, a WAF rollout from report mode to block mode, security headers and a Content Security Policy with ScriptAttest. In Claude Code, run /plugin marketplace add quantcdn/agent-skills and then /plugin install quant@quantcdn. For QuantCode and other agents, copy the skill folders from the quantcdn/agent-skills repository on GitHub into the agent's skills folder. A new getting-started guide in the Quant documentation, Use Quant with AI coding agents, explains the installation. These skills are files your coding agent reads, separate from the AI skills you create in the dashboard for Quant AI agents. They are free to use under the MIT licence.
A new AI audit log page in the AI menu lets organisation owners and admins search the organisation's AI audit records. The records show who used which model and tool. You can filter a date range of up to 31 days by user, API token, model, record type, decision, tool, outcome or interface. You can open a single record and download the results as JSON Lines. New v3 API endpoints return the same records to tokens with the new ai:audit:read scope, which full-access tokens include. If you have AI governance permission, you can also export a date range from the Export audit log card on the Governance page. You can read records from the last 365 days, and every search, record view and download is itself recorded in the log.
New AI platform
Slack bot messaging API
Your own tools can now read and post in Slack through a Quant Slack bot. The API endpoints are under /api/v3/organizations/{org}/ai/slack-bots. In the Slack bot editor, turn on API access, choose the bot's API channels from its allowed channels, and set a limit of 1 to 60 writes per minute. A token with the ai:slack scope can then list the bots it can use and read recent messages and threads in those channels. It can also post messages, and upload PNG, JPEG, GIF or WebP images of up to 4 MiB with an optional caption, title and alt text. Each post includes a line that names the person who sent it. An optional Idempotency-Key header stops a retried request from posting twice. Full-access tokens do not include ai:slack, so create a token with that scope. The token's owner needs a verified email address. The Slack app manifest now also asks for the file upload permission that image posts need.
New AI platform
Spend warning emails
When your organisation's AI spend reaches the warning threshold set in AI governance, everyone with AI governance permission now receives an email, at most once a month. The email shows spend against the monthly budget and says whether requests are already blocked because the budget is reached. Leave the threshold empty if you do not want the email.
New AI platform
Claude Haiku 5.5
Claude Haiku 5.5 is now listed in the AI API, with Australian-only inference for government and public organisations. It has a 200K-token context window with a 1M-token option and up to 128K output tokens, and it supports tools, vision, streaming and structured JSON output.
New CLI
Run a built site locally with quant serve
CLI v7.1 adds quant serve, which runs a built site on your machine and needs no API token. It serves the site with its edge functions (request, auth and content filter), edge rules and custom 404 page. It can start your development server and proxy to it, including WebSocket connections, or rebuild the site when files change. A local KV store is seeded from your project's KV declarations and fixtures, and quant kv --local reads and writes it while the server runs. Version 7.1 also adds quant kv-indexes, which declares the KV indexes listed in quant/meta.json on your project's KV stores in Quant. It adds only the indexes that do not exist yet and does not change existing ones. Version 7.2 runs KV ledgers locally, with a manual clock for testing time windows. In version 7.2, quant serve also replaces inline function tokens in pages and applies project headers, basic authentication and trailing-slash redirects from a local quant/serve.json.
New Cloud CLI
Environment variables with qc env vars
Cloud CLI v0.7.0 adds qc env vars list, get, set and unset, so you can read and change a Quant Cloud environment's variables from the terminal. set can read a value from standard input, which keeps it out of your shell history, or merge a .env file without removing variables that are not in it. list hides values unless you pass --show-values. Changes take effect on the environment's next redeploy, which you can start with qc env state redeploy. Secrets are still managed in the dashboard.
The crawler create and update API endpoints now accept browser_config settings for browser-mode crawls: capture_api_responses, wait_for_network_idle and use_rendered_html. With capture_api_responses, a crawl stores the JSON responses that a page loads in the browser, so a static copy of a site whose navigation or content is built from a JSON endpoint shows that content. You can send the settings at the top level of the request or as a nested browser_config object, and the Terraform quant_crawler resource has a matching browser_config block. The dashboard crawler form does not show these settings.
New Domains & DNS
DNS API
You can now read your organisation's DNS zones and records, and create A records, through the API at /api/v2/organizations/{org}/dns/zones. Access uses two new token scopes, dns:read and dns:write, together with the token owner's domain permissions, and the token must not be limited to individual projects. Send record names relative to the zone, with @ for the apex; a name that already ends in the zone's domain is refused with a clear error. The read-only token preset now includes dns:read; existing tokens are not changed. Other record types, changes and deletions remain in the dashboard.
The new @quantcdn/edge-conformance npm package holds 153 test cases that describe how the Quant edge responds to a request. The cases cover rule matching, static paths and the custom 404 page, and edge functions, auth functions and content filters. They also cover inline functions and the project settings for custom headers, basic authentication and trailing-slash redirects. From CLI v7.1, Quant tests quant serve against these cases, and quant serve --version-info shows the suite version that the CLI was tested with. You do not need to install the package to use quant serve. Each case states one request and the expected response, so they are also exact examples, such as which paths a wildcard rule pattern matches. To test your own tool that copies Quant edge behaviour, write an adapter file for it and run npx edge-conformance --adapter <file>. It needs Node.js 20 or later and is free under the Apache-2.0 licence.
New KV stores
Format 2 stores
A new store format keeps keys in an index, so reads are strongly consistent and an edge function can list one collection of keys without scanning the whole store. A key's collection is the text before its last :, so the key customer:abc:world:w1 is in the collection customer:abc:world. In edge functions, kvStore.getEntry() returns a value with its version and timestamps, and getMany() reads up to 100 keys in one call. Writes can be conditional with ifAbsent or ifVersion, and deletes with ifVersion. atomic() applies a transaction of up to 25 operations completely or not at all. With these, you can create a record only once, or update several keys together without overwriting another writer's change. Failures throw a KVError with a code. Format 2 is the default for stores created in the dashboard, and stores created through the API stay format 1 unless the request asks for format 2. To upgrade an existing store, use the store page: it runs a dry-run check first, and you can roll back for 14 days.
Use ttl instead of expiration or expiresIn in edge function writes, because writes that pass those options now fail.
Format 2 stores can keep indexes that you declare on a collection. An index lists a collection by the values of up to three fields in a sort order you choose. A unique index refuses a write that would give two keys the same value, for example one account for each email address, and getBy() looks a record up by that value. A count index keeps a running count for each group, which count() returns in one call. Existing items are indexed before an index becomes active. Declare indexes on the store's Indexes tab, which shows the extra write cost of each design, through the API, or in your repository's quant/meta.json. A store can have up to 20 indexes, with up to 5 on one collection.
A ledger is an append-only collection in a format 2 store for records that must stay as written, such as audit trails, registers of AI use, and consent or approval logs. Edge functions add entries with kvStore.append(), and you can also add entries from the dashboard. Nobody can change or delete an entry afterwards, and each entry records which writer added it. Entries are copied to a locked archive for the retention period you set, optionally with an external anchor in your own storage bucket or with a timestamp authority. Once a day, the anchor sends a hash of the store's latest signed digest to a witness outside Quant, so you can prove that the entries it covers did not change. From a store's Ledgers tab you can declare a ledger, browse entries, and run an integrity check and keep the signed report. On the same tab, you can prove that a record was on the ledger with a proof statement for auditors. Fields you mark as personal stay hidden until someone with permission reveals them. An erasure request makes one person's personal data unreadable in every entry and gives a signed receipt, while the ledger stays verifiable. You can check reports, proofs and receipts offline with the open-source Quant ledger verifier.
A store's page now shows its item count, size, daily reads and writes for the last 30 days, and an estimated monthly cost. A new Export tab downloads every item as compressed JSON Lines files and shows the cost of an export before you start it. An export reads a point-in-time snapshot, so it uses none of the store's read capacity. The files are kept for 7 days, and the Export tab gives new download links while they are kept. Exports are limited to stores of up to 1 GiB, and ledger fields marked as personal are exported as placeholders.
KV stores and ledgers now have their own role permissions instead of sharing the content permissions: read, write, delete and administer for stores, and read, append, administer, reveal and erase for ledgers. Two new optional organisation roles suit records and privacy staff: Records Officer can read stores and ledgers, and Privacy Officer can also reveal personal fields and carry out erasure requests.
Test edge functions locally with @quantcdn/edge-kv
The new @quantcdn/edge-kv package on npm gives you the edge kvStore API and a local store on your own computer, with no Quant project or API token. Use it to run and test edge functions that read and write KV stores, for example in unit tests. It checks arguments and throws KVError codes as the edge runtime does, and it includes the runtime's TypeScript types. It supports format 1 and format 2 stores, indexes, transactions and ledgers. You can also run a ledger erasure request locally, which returns a receipt signed with a local key. You can hold an index in its building or failed state, and you can set the clock. These controls let you test how a function handles an index that is not ready or a ledger time window. quant serve in CLI v7.1 and later uses the package for its local KV store. The package is open source under the Apache 2.0 licence and needs Node.js 20 or later. It is for development and tests only, so do not store real secrets or personal data in it.
New Projects
Content API token rotation
You can now rotate a static project's content API token from the project's integration settings. You choose whether the old token stops working immediately, after 24 hours or after 7 days, so pipelines that still use it keep deploying during the changeover. Requests made with the old token during that time receive a Quant-Token-Deprecated response header. If the project deploys from a linked GitHub repository or a Studio environment, the QUANT_TOKEN secret there is updated for you. Update the token anywhere else you use it, such as the Drupal module, WordPress plugin, CLI configuration or crawler webhooks.
New Quant Cloud
Application metadata
You can now attach up to 50 key and value pairs, such as a team or cost centre, to a Quant Cloud application. Use the new Metadata section on the application page or the v3 API (PUT or PATCH .../applications/{application}/metadata). The applications list shows metadata as labels that you can click to filter, and its search matches names, labels or key=value. The API filters the list with metadata[key]=value, and the Terraform quant_application resource manages metadata as a map. Metadata is visible to everyone who can see the application, so do not store secrets in it.
New Quant Cloud
Slow database queries
The Database tab on an environment's metrics page now lists queries that took longer than one second in the selected time range. Queries are grouped by shape with their values replaced, and each group shows how often it ran and its average and maximum time. The v3 API returns the same data at the environment's slow-queries endpoint. If slow query logging is off for the database, the tab says so.
New QuantCode
Claude Code settings, hooks and memory
Repositories set up for Claude Code now work the same way in QuantCode. QuantCode applies the allow, ask and deny permission rules in .claude/settings.json, .claude/settings.local.json and your user settings, and runs SessionStart, UserPromptSubmit and Stop command hooks. A hook that comes with a project asks once before it first runs. QuantCode also loads the project memory that Claude Code keeps and can save new memories in the same format. By default, QuantCode now asks before it opens a shell in a running cloud container. Hooks do not run in autonomous jobs or sub-agent sessions.
New QuantCode
Live console for autonomous jobs
Each QuantCode Agent job in the dashboard now opens a console that streams the session while it runs. Each tool call appears as a titled card with its duration. The console also shows the job's running cost and links to any pull requests that the agent opens. You can send the agent a message, interrupt or pause it, or cancel the job. When a job must ask before an action, an Approval needed card shows the request. You can approve it once, always allow it for the session, or deny it. A finished session replays in full and is marked as ended. Jobs started from Slack also post a status message for the session in the Slack thread.
New QuantCode
Scheduled and repeating prompts
You can now ask QuantCode to run a prompt later or repeatedly, for example "check the build in two minutes" or "from 1am tomorrow, check the deployment every minute for an hour". A schedule has a start time, an end time or a run limit, and an interval or a cron expression with a time zone. It runs in the session that created it, on that session's current model, while the project is open. The desktop app's Schedules button and the terminal interface list each schedule with its next run, with controls to pause, resume, complete or cancel it.
New Rules
Manage edge rules in code
You can now keep a project's edge function, auth function and content filter rules in an edge-rules.json manifest in your repository. Deploy them with the CLI's quant rules command (v6.3 or later) or the deploy action, using a project-scoped API token. A dry run previews the changes first, and rules can refer to functions by the id in your functions manifest. Rules deployed this way show a Managed in code badge and are read-only in the dashboard and the Rules API. If such a rule was changed outside code since the last deploy, the deploy writes nothing and lists the changed fields. A forced deploy overwrites them. A deploy moves only code-managed rules, and places each one among the other rules by its weight. Rules added in the dashboard keep their order, and Delete all rules and Clone from project leave code-managed rules in place. Code-managed rules can use negative weights to run before any other rule.
Auth function rules and more conditions in the API
A new rules/auth-function endpoint in the v2 Rules API lists, creates, updates and deletes rules that run an auth function, identified by the function's UUID. The existing rules/auth endpoint now lists only basic authentication rules. Rules created or updated through the API also accept the asn, ua (user agent) and traffic_source conditions that the rule editor offers, and a PATCH that leaves out a condition keeps its current value. API version 4.26.0 documents the new fields and the existing cookie_name field, which applies a rule only when that cookie is present in the request.
Use the rules/auth-function endpoint for auth function rules, because rules/auth now lists only basic authentication rules.
New Search
AI answer safety level
The AI Search settings tab now shows the answer safety level that applies to AI answers across your organisation (OFFICIAL, OFFICIAL:Sensitive or PROTECTED). It also shows whether the level was set for search, inherited from your AI governance level, or is the default. Search uses search-specific versions of these levels, so links and contact details from your pages are not removed from answers. If you have AI governance permission, you can set a level for search without changing AI governance.
Sentinel 2.1 can run a whole scan on your own OpenAI-compatible inference endpoint instead of the AI services in your Quant organisation. Set LLM_PROVIDER=openai with LLM_BASE_URL, LLM_API_KEY and LLM_MODEL. In this mode, Sentinel does not need Quant credentials. If you do not set these variables, Sentinel uses your Quant organisation as before.
New Studio
Content author, publisher and manager roles
Three Studio roles now separate writing from publishing. Studio Content Authors can edit and delete content and publish preview branches, and when they open or save on the production branch, Studio moves their work to their own working branch. Studio Content Publishers can also merge reviews, publish, deploy and write to the production branch. Studio Managers can also manage branches and Studio settings. The person who opened a review cannot merge it, and controls that a user cannot use are hidden. Existing Studio Editor users become Studio Content Publishers with the same access.
New Studio
Orbit edge SaaS template
The new Orbit template is a working project-management dashboard. It runs on the CDN with one edge function and a KV store, and it needs no application server or database. When you create a site from it, the dashboard sets up the KV store, the edge function and its rules. You can edit its content in Studio, but you do not have to.
New Terraform
KV indexes and ledgers
New quant_kv_index and quant_kv_ledger resources manage indexes and ledgers on format 2 KV stores. New quant_kv_ledger and quant_kv_ledger_public_key data sources read a ledger's state, retention and last verification, and the public key that verifies its signatures. The quant_kv_store resource gains a format argument that is set when the store is created, and a plan that would change the format of an existing store fails instead of replacing it. A ledger's definition cannot change, except that its retention can be raised. Destroying a ledger fails while its retention period runs and never deletes its archive. The Pulumi provider maps the same resources.
Built-in filter policies that are off by default now appear in the agent editor and on the Governance page, even when they are off for the organisation. They are labelled as off organisation-wide, and you can now apply one to a single agent or add it to the mandatory list.
Improved AI platform
Stop a running workflow
The workflow editor now has a Stop button on running and paused workflow runs. If a run cannot be stopped, for example because it has already finished or runs inline, the editor says why instead of showing a generic error.
Improved AI platform
Australian-resident models only on QuantGov
On QuantGov, AI model lists now show only models that run in Australia, so your AI requests stay in Australia. This applies to the model pickers and the models API. The model list in AI governance shows only these models, and an allowlist or blocklist cannot add others.
When the AI API refuses a model outside your region, it returns the error code model_not_available_in_region. AI pages in the dashboard now say that the model is not available in this region and ask you to select a different model.
Improved API
Rate limits per token
API rate limits now apply to each API token and organisation instead of each client IP address, so applications and CI runners that share an outgoing IP address no longer share one limit. The default is 240 requests per minute. Responses include RateLimit-Limit, RateLimit-Remaining and RateLimit-Reset headers alongside the X-RateLimit-* headers, a 429 response includes Retry-After, and the limits are now documented in the API reference.
Improved CDN
Redirect caching
Redirect responses now carry a Cache-Control header that tells browsers to keep a redirect for five minutes, so returning visitors follow a changed redirect soon after you change it. At the edge, redirects stored as project content, such as those created with quant redirect, are cached for 30 days and purged automatically when they change. Redirect rules are cached at the edge for one hour and are not purged when you edit the rule, so purge the affected paths if a change must apply at once. A Cache-Control header that another rule sets is kept.
Improved CLI
v7 requires a UUID for every edge function
Quant CLI 7.0 is a breaking release for edge functions only: every entry in a functions manifest needs a uuid, and quant function, quant auth and quant filter need the UUID argument. Earlier versions derived a UUID when one was missing, and the single-function commands created a new function on every run. The new quant functions-add command adds an entry with a new UUID, and the whole manifest is checked before anything is uploaded. The README's upgrade notes explain how to keep the UUIDs version 6 derived.
Add a uuid to every entry in your functions manifest before you upgrade to Quant CLI 7.
Browser-mode crawls now download each asset from your origin once instead of twice. The crawl delay now paces page requests only, which are the requests that reach your CMS and database, so images, CSS and fonts are no longer held back by it.
Improved Dashboard
Keyboard and screen reader access
Dashboard pages now have a skip link and a main landmark, and icon-only buttons and links have accessible names, with a tooltip where it helps. Confirmation dialogs are announced as dialogs, put focus on Cancel first for destructive actions and return focus to the button that opened them. Table headers and checkboxes now work from the keyboard, and the skip link and status colours have better contrast, including in dark mode.
Improved Dashboard
Time zone setting on your profile
You can now choose the time zone that the dashboard uses for times such as deployment, backup and content times. By default, these times show in your browser's time zone. To use a different time zone, select it under Time zone on your profile page. To go back to the default, choose Use my browser's time zone.
Improved GitHub Actions
Deploy action edge rules, KV indexes and v7
The deploy action can now deploy code-managed edge rules after functions and assets, in v6 and v7, with new rules, api-token, api-base-url and rules-force inputs. The rules step uses a separate project-scoped API token, not the content token, and checks access and changes with a dry run before anything is uploaded. Version 7 runs Quant CLI 7, so every entry in the functions manifest needs an explicit uuid. Version 7.1 adds a kv-indexes input that declares the KV indexes listed in quant/meta.json on your project's KV stores before the functions deploy. It adds only the indexes that do not exist yet. Workflows pinned to v6 keep Quant CLI 6.3.
Add a uuid to every entry in your functions manifest before you move a workflow to deploy action v7.
The v2 KV API now covers format 2 stores: versioned entries with conditional writes, atomic increments, batch reads of up to 100 keys and transactions of up to 25 operations. It also covers cursor-based key listing, index declaration, lookups and counts, format upgrades, 30-day usage with a cost estimate, and store export. Ledger endpoints declare ledgers, append and read entries, verify a ledger, return proofs and signed reports, and preview and carry out the erasure of a person's data with a signed receipt. Ledger administration and erasure need a token with the kv:ledger:admin or kv:ledger:erase scope selected explicitly, and full-access tokens do not include them. Creating a store through the API now waits up to 20 seconds and returns status: ready. If the store needs longer, the request returns 202 with status: provisioning and a Retry-After header. A refused store delete returns a code that says what blocked it.
Improved KV stores
Confirm before replacing an existing key
Creating an item in the dashboard with a key that already exists now asks before it replaces the stored value, instead of replacing it without a warning. When a plain value would replace a secret, the dialogue says so and the button reads Replace secret. Keep existing returns you to the form with nothing changed.
Improved Projects
Templates set up KV stores, rules and settings
Creating a static site or Studio site from a template now sets up everything the template declares in its quant/ folder. This includes KV stores with starting values and indexes, edge functions and the rules that use them, and project settings such as inline functions. A template can generate random secrets during setup, and a one-time credential created this way is shown to you once when setup finishes. Setup checks the template's declarations before it creates anything, so a broken template stops at setup with a message instead of failing at its first deploy. Template KV stores are created in format 2, and the GitHub workflows that setup generates use deploy action v7. The start-from-template guide now lists every template with its framework, and a new documentation page explains the files a template can contain and how to write your own.
Deleting a project now also deletes the KV stores that belong to it, while stores that are only linked to the project are unlinked and kept. If one of the project's stores holds a ledger, the delete is refused and names the store, because ledger entries must be kept for their retention period. The same check applies to deletes through the API and to cancelling a subscription.
Improved Quant Cloud
Builds match each environment's architecture
Applications created from a starter kit or imported from GitHub now build their container images for the CPU architecture of each environment, x86_64 or ARM64. Creating an environment or changing its architecture updates the linked repository's build settings and tells you which architecture the next build targets. New applications from templates, starter kits and GitHub imports now default to x86_64. New documentation explains build architectures and how your own ARM64 runner speeds up ARM64 builds.
On an environment's Env & Secrets tab, saving a variable under the name of an existing variable, or a secret under the name of an existing secret, now asks before it replaces the value. Keep variable or Keep secret leaves the current value in place. A variable and a secret can no longer have the same name. The tab now refuses a name that the other kind already uses, and bulk edit mode and the v3 variables API refuse a variable with the name of an existing secret. If a name is already used by both, the tab shows a warning that lists the name, so that you can delete either the variable or the secret.
Improved QuantCode
Autonomous jobs use a model suited to each agent
Autonomous jobs now run most agents on Claude Sonnet 5.5 and the review check on Claude Haiku 5.5, where your AI governance settings allow them, instead of Claude Opus 5.5 for every agent. This lowers the cost of a typical job. Security and supply-chain review agents keep their configured model, and the job report breaks down spend by agent and model. Before a job finishes, code changes that have not been reviewed in their final state are reviewed, and the run summary lists changes from the repository's commits. Model overrides in a job's configuration now apply to any agent, not only the orchestrator, and long jobs keep their GitHub access until the final push, or report a push that fails. A job no longer starts without its tool catalogue, so the agent always has web search.
Improved QuantCode
Automatic updates for the desktop app and CLI
The macOS desktop app now checks for updates 10 seconds after it starts and every 4 hours, downloads a new version and asks you to restart to install it. Homebrew installs of the CLI with automatic updates on now install any newer release, not only patch releases, never move to an older version, and update only the copy that Homebrew installed. Other installs now show a notice when a new version is available, and quantcode upgrade explains how to reinstall.
Improved QuantCode
Smaller requests and steadier prompt caching
Each request now lists your local skills once instead of twice, which removes about 4,600 tokens from every request when 45 skills are installed. The system prompt also stays the same across sessions in a repository, so cached prompts can be reused. When two skill folders contain a skill with the same name, QuantCode now picks the same copy every time.
Improved Rules
Function and domain checks when you save a rule
When you create an edge function, content filter or auth function rule, or change the function it runs, the rule editor and the Rules API now check the function. It must be a published function of a type the rule can run, and the error says why if it is not. Rule domains are exact host names, such as www.example.com. The Rules API and edge rule manifests do not accept wildcard domains. Deleting an edge function now lists the rules in the project that use it.
Improved SDKs
Format 2 KV stores, ledgers and application metadata
The TypeScript, PHP, Python and Go clients are now generated from API v4.29.0, and the Go client also shipped v4.25.0 earlier in the month. New coverage includes format 2 KV stores and ledgers (versioned entries, transactions, indexes, export, upgrades and usage), application metadata and the Slack bot messaging API. It also covers auth function rules, the new rule conditions, and crawler browser settings. The TypeScript client moves from v4.21 to v4.29.0. Version 4.29.0 also removes the image generation methods and models, because the image generation endpoint was retired this month.
Improved Search
Search settings in the projects API
The v2 projects API now returns search_enabled and search_config for each project and accepts search_config changes. A change is merged one section at a time, so a PATCH that changes only the extractors keeps the other search settings, and tools that manage projects as code can read these settings back.
Improved Studio
Site migration into the CivicTheme template
The site migration tool, introduced in February, can now build into the CivicTheme template instead of generating a bespoke site: choose it under Build as when you start a migration. Pages are rewritten into CivicTheme components and keep their original paths. The site name, menus, logo, colours and fonts come from the source site, and colours are adjusted to meet contrast requirements. Before the site is assembled, the review step lists each repeated content pattern and how it will be mapped, and you can change the mapping.
Improved Studio
Publish applies rules, KV stores and project settings
The edge functions and forms as code introduced in August now extend to the rest of your site's quant/ declarations. With Quant Builder on, a Studio publish also applies your edge rules, KV stores with their declared values and indexes, and project settings. A site that publishes from Studio therefore needs no separate GitHub deploy. Each branch environment gets its own KV store. Each declared value is written only once, so later publishes do not undo edits or deletions made to that value. The Deployments page shows the result of each step, including indexes that are still building or have failed.
Improved Studio
Faster loading for uncached projects
Projects and branches that are not yet cached now open faster in Studio, including when several people open the same project at the same time. Saves and site migrations now also retry automatically when a write to GitHub fails because of a dropped connection or a temporary GitHub server error, after checking whether the first attempt succeeded.
Improved Studio
Site migrations start from the main navigation
Site migrations now choose pages from the source site's navigation first and then from its sitemap, including sitemap indexes, so a migration with a page limit starts with the site's main sections. A crawl that stops making progress ends automatically so that the migration continues, and a temporary network error while the new project is created is retried instead of failing the job. Bespoke sites now also include the source site's footer menu. A bespoke migration now checks the new site's deployment build before it is marked complete, and first tries to repair a failing build before it reports a failure.
Improved Studio
Grid cards in the component editor
In the component editor introduced in September, a grid's Contents list now comes first and lets you move, duplicate or remove cards. Clicking a card in the preview opens it in the editor instead of following its link.
Improved Studio
Keyboard and screen reader access
You can now open files and folders in the Studio file tree from the keyboard. Shift+F10 or the context menu key opens a row's menu, which you move through with the arrow keys and close with Escape. The AI assistant's conversation now scrolls from the keyboard. The assistant, content list, schema and media controls and the preview frames have accessible names, and code editor comments, line numbers and hint text now meet the WCAG 4.5:1 contrast ratio.
Improved Terraform
User agent, traffic source and cookie name rule conditions
From provider v5.15.0, every rule resource has user agent (ua), traffic source and cookie name conditions, sent to the API and read back. The auth, bot challenge, header and serve static rule resources also send their asn condition, so all rule resources support it.
Adding a subscription or changing your search plan with a card that needs extra confirmation from your bank did not always take you to the confirmation step, and could end in an error page. The dashboard now takes you to the payment confirmation page, and when you finish there it returns you to your subscription page with a notice while the payment is confirmed.
Fixed Account
Cancel subscription button
The subscription page no longer showed a Cancel subscription button. The button is back for subscriptions that can be cancelled online, and it opens a confirmation dialogue in which you type a confirmation phrase before the subscription is cancelled.
Fixed AI platform
Streaming usage and reasoning effort in the OpenAI-compatible API
Streamed /oai/v1/chat/completions requests that set stream_options.include_usage did not receive the usage chunk, and the reasoning_effort parameter was ignored. Both now work, so you can get token usage for streamed responses and reduce or turn off reasoning, for example on Claude Haiku 5.5, with none, minimal, low, medium or high.
Fixed AI platform
Slack bot channel and user search
In the Slack bot editor, a channel or user search that failed, for example because Slack limited requests, showed an empty list. The workspace then seemed empty. The search now shows a message when it fails, and shows No options found when nothing matches.
Fixed CDN
Purging a list of URLs through the API
Sending a urls array to the v2 project purge endpoint returned an error. Each URL in the array is now purged, up to 100 per request. Every URL is checked before anything is purged, and if some paths or the cache keys cannot be purged, the response is an error that names them, not a success. The API reference now states that cache_keys is optional with urls and that purges are soft unless soft is false.
Fixed CDN
Stored files requested with a query string
A file that Quant holds, such as a Drupal image style, was fetched from your origin when the request included a query string, for example ?itok=. This happened on paths that a proxy on 404 rule covers. Quant now serves its stored copy when the published item is a file. Pages, redirects and unpublished items that are requested with a query string still go to your origin.
Fixed Cloud CLI
Platform selection with --platform
The --platform flag and the platform key in .quant.yml had no effect, so commands always ran against the active platform (QuantCDN or QuantGov Cloud). From v0.7.1, they select another platform you are signed in to for that command, without changing the active platform. Separately, since v0.7.0, commands that run in CI or with piped output no longer print spinner frames.
Fixed Crawler
Assets that page scripts request
Browser-mode crawls could publish pages without images and other files that the page's JavaScript requests after it loads, including on client-rendered sites and on pages that depend on a script library from another domain. Crawls now wait until each page's network activity settles, up to a limit that scales with the crawl delay, and let third-party scripts load while the page renders. Analytics, advertising and session-recording services are still blocked, and third-party files are not stored in your static copy.
Fixed Crawler
Scheduled crawls no longer overlap a running crawl
A scheduled crawl could start while an earlier run of the same crawler was still running or stopping, so two runs crawled your site at the same time. A scheduled start is now skipped while another run is active or stopping. Starting a crawl through the API while the same crawler has a run in progress now returns 409 Conflict. A crawl started from the dashboard or the API is also refused until a stopping run has finished. Runs that are still active after five days are stopped and shown as timed out in the run list.
Fixed Dashboard
Editors keep saved content that fails to load
If the saved version of a content page or an edge function could not be loaded, the code editor or function editor opened empty, so a save could replace the content with a blank version. Both editors now show an error with a Retry button until the content loads, and the custom headers editor does not save until your existing headers have loaded.
Fixed Dashboard
Missing icons, broken links and template names
Many buttons across the dashboard, including on the crawler, project creation and Quant Cloud forms, were missing their icons, and the icons now show. Back and Cancel on the new organisation page led to a missing page and now return to the dashboard home. The Jamstack starter kit wizard's Back to options, Create another and Start over links went to the dashboard home and now return to the static site pages. Template galleries now spell names such as WordPress, Drupal CMS and Next.js correctly. Dates on the container image security and GitHub connection screens and the one-day metrics chart show the month as a word, so they read the same in every locale.
Fixed Domains & DNS
DNS records page load errors
When a zone's records could not be loaded, the records page showed no message, and it now shows the error.
Fixed GitHub Actions
Quant Cloud action fixes
Deploying with the Quant Cloud environment action reset the environment's scaling to a minimum and maximum of one task. From v4.1.0, the action leaves scaling as it is, so values set in the dashboard are kept, and new optional min_capacity and max_capacity inputs set scaling from the workflow. From v1.0.3, the environment variable action no longer fails when base_url ends in /api/v3. The environment sync action v4.1.0 also gains a fail_on_skip input that fails the step when the platform refuses a sync, for example for an environment with no database, instead of passing with a warning.
Update the Quant Cloud environment action to v4.1.0 or later so that deploys keep the scaling set in the dashboard.
Fixed KV stores
New stores appear after you create them
A KV store created from the dashboard did not appear in the list until you reloaded the page. New stores now show at once, and the dashboard tells you if a store was created but could not be added to the project.
Fixed Quant Cloud
Creating applications from templates, starter kits and GitHub
Applications deployed from a pre-built template that needs generated secrets could not start, because the secrets were not created, and they are now generated during setup. Applications built from a template's repository now deploy your own build instead of the template's published image, and new starter kit repositories no longer include the template's own deploy workflow. Importing an application from GitHub with a branch other than main created a build workflow that ran only on pushes to main. The workflow now runs on the branch you choose, and its preview shows the exact file that is committed.
Fixed Quant Cloud
Environment settings and database charts
Saving a changed target value on an existing autoscaling policy in the environment edit form did not apply the change, and the new target is now saved. The environment edit page showed an error page when the environment's details could not be loaded, and it now shows the error message on the page. The database IOPS, latency and throughput charts listed each read and write series three times in the legend, and each series now appears once, as its average.
Fixed QuantCode
Interrupted cloud tool calls are not repeated
If the connection dropped while a cloud tool, such as web search, was running, QuantCode could send the request again, so the tool could run twice. QuantCode now stops with a message that the call was not sent again, and asks you to check its result in the dashboard. It keeps the interrupted call in the session, so the agent knows the tool started. Web search calls also keep their query in the session history.
Fixed QuantCode
Stability fixes
An image whose file extension did not match its contents, such as a JPEG saved as .png, was refused and made every later turn in the session fail. Images are now sent with their actual type, and affected sessions recover. Language servers, such as the TypeScript language server, could fail to install on macOS when the cache path included a symbolic link. Under zsh, shell commands failed on words that start with =, such as ==, and on wildcards that matched nothing. An agent that took over through a handoff could also stop at its step limit straight away. These issues are now all resolved.
Fixed Rules
Redirect status and static error page in the API
A redirect rule created through the API without redirect_code was saved without a redirect status code, and new redirect rules now default to 301. Reading a proxy rule through the v2 Rules API returned an empty static_error_page value even when the static error page was on. Tools that read the rule back, such as Terraform and Pulumi, saw it as off. It now reads back as true. The Rules API reference is also corrected, for example to state that origin_timeout is in milliseconds and that inject_headers are added to the origin request.
Fixed Studio
Simultaneous publishes
When two people published the same site at the same time, the live site could lose its styles. Studio now runs one full publish per site at a time. A second publish stops before it builds and shows who is publishing and when the site will be free, and a publish left running in a closed tab is released after two minutes.
Fixed Studio
Publishing reliability
A publish with the Force option on the Deployments page no longer downloads every media file again, and it now restores pages that were unpublished outside Studio. A media upload failure no longer stops pages and their assets from publishing; the error now appears on the Deployments page. The Deployments page no longer times out when GitHub responds slowly, and page counts on publish runs are accurate. After a merge, Studio names the workflow that runs and shows it as running, passed or failed, rather than as a deploy. A full publish also no longer removes a live page whose address differs only in letter case from a page in the build.
Fixed Studio
Branches with a slash in their name
On branches with a slash in their name, such as feature/new-page, Studio could not open files, build and publish the site, or show media in the preview. These now work.
Fixed Studio
Schema editor, History page and error messages
The schema editor did not open for a collection with a field whose default value is a list or an object, such as an empty array, and these defaults now show as JSON. When GitHub could not be reached, Studio showed raw technical error text, and Build & Publish or Re-index could stop with an error dialogue. Studio now shows a plain message with a next step, and Build & Publish and Re-index stop before they begin and offer Retry. The History page no longer says a branch has no commits when its history failed to load. A link that opens an entry in the visual editor now opens that entry. Connect GitHub on the template wizard now returns you to step 2 with the organisation picker open.
Fixed Terraform
Deleting environments with their application
Destroying an application and its environments in one run could fail with an error saying that non-production environments must be deleted first. The provider now waits until each environment is removed before it continues.