New AI platform APIs for tasks, file storage and asynchronous chat, a static site wizard that imports from GitHub, new CDN metrics endpoints, vector database search for Drupal, and the launch of ScriptAttest for Content Security Policy management and monitoring.
A new Tasks API is available in the AI platform for managing multi-agent workflows. You can create, assign, update and delete tasks, organise them into task lists, and build dependency graphs between tasks. The list endpoint can filter tasks by status, assigned agent and task list. It also has a reverse dependency lookup (dependsOn) that finds all tasks waiting on a given task. To remove a task and all the tasks that depend on it, add the query parameter cascade=true to the delete request.
New endpoints let you upload, list, retrieve and delete files in your organisation's AI file storage, such as generated documents, images or data files. You can upload files up to 7 MB directly in the request or request an upload URL for larger files, attach your own metadata, and filter the file list by metadata with cursor pagination.
Chat inference and agent chat requests can now run asynchronously for long-running work. Set async: true, and the API responds at once with a request ID and a URL. Poll that URL for the status and the result. If the model calls a tool that your application runs, the execution pauses until you send the tool result to the callback endpoint (POST /api/v3/organizations/{org}/ai/chat/callback). Synchronous and streaming requests can also run for longer than before.
Three new endpoints have been added to the CDN metrics v2 API under /v2/organizations/{org}/projects/{project}/metrics/: hourly (minute-by-minute data for the past hour), daily (day-by-day data), and monthly (month-by-month data). Each endpoint supports optional domain filtering, configurable metric selection (hits, bytes, and others), and ISO 8601 or Unix timestamp formats.
New Dashboard
Static site wizard with GitHub import
Creating a static site now starts a wizard. You can start from a template, import an existing GitHub repository, or set up the project manually. When you import a repository, the wizard detects its build settings and its framework: Next.js, Nuxt, Astro, Hugo, Gatsby, Jekyll, Eleventy or Docusaurus. The wizard then creates the project, adds the deployment secrets and variables to the repository, and opens a pull request with a GitHub Actions deploy workflow. After you merge it, each push to your chosen branch deploys the site.
The Quant Cloud AI provider module for Drupal now includes a vector database integration (ai_provider_quant_cloud_vdb) compatible with the Drupal AI Search module. You can create Search API servers backed by Quant Cloud's vector database and index Drupal content for semantic search. A direct PHP client is also available for use outside of Search API. When indexing, you can have the Quant Cloud platform generate the embeddings instead of your Drupal site, and documents are uploaded in batches for faster indexing.
New ScriptAttest
CSP management service launch
ScriptAttest (scriptattest.com) is a new hosted service that builds, tests and monitors a Content Security Policy (CSP) for your site. It has its own sign-up and accounts, separate from the Quant dashboard. Discovery scans load the pages you add in a real browser and list the scripts, stylesheets, fonts, images and connections they use. You approve each source by domain, or approve inline scripts and styles by hash, and ScriptAttest generates a strict policy from your approvals. The source list flags public package CDNs as a supply-chain risk. A validation scan tests the policy against your live pages before you deploy it. Each generated policy includes a reporting endpoint for your site, so browsers send violation reports to ScriptAttest. Email alerts tell you about new violations. On the Violations tab, you can approve a reported resource, which adds it to a new policy version, or dismiss the report. The Security Headers tab grades headers such as HSTS, X-Frame-Options and Referrer-Policy. In Settings, you can invite team members as admins or members, and you can switch between the organisations you belong to.
New ScriptAttest
Script attestation and drift detection
Attestation scans record each script on a page with its source, its content hash and the domains it sends requests to. The first scan becomes your trusted baseline. Later scans report scripts added, changed or removed since the baseline, rated by severity. When a change is expected, such as a vendor update, you accept that scan as the new baseline. From an attestation report, you can mark domains whose scripts change by design, such as an internal CDN or an A/B testing service, so their script changes show as expected. The ScriptAttest tab also lists Subresource Integrity (SRI) values for external scripts that do not change, ready to copy into integrity attributes. On the Schedules tab, you can run attestation and policy validation scans daily, weekly, monthly or on a custom cron expression.
New ScriptAttest
CSP deployment to QuantCDN and QuantGov projects
You can link a ScriptAttest site to a QuantCDN or QuantGov project from the CSP Management tab. You sign in with your Quant account and choose the project. Deploy to QuantCDN then adds the policy and its Reporting-Endpoints header to the project's Custom headers page, and keeps your other custom headers. When approving a violation publishes a new policy version, ScriptAttest deploys that version to the linked project automatically. Organisation owners and admins get an email when an approval publishes a new policy version, and another if the deployment fails. The CSP Management tab shows the policy version and time of the last deployment.
The chat inference and streaming endpoints now accept OpenAI field names alongside the existing format. They accept model for modelId, max_tokens for maxTokens and stop for stopSequences, and the full OpenAI tool-calling format, including tools, tool_calls and tool role messages. Existing OpenAI integrations can move to the Quant Cloud AI platform without rewriting payloads.
Improved AI platform
Vector database query modes, metadata filters and document deletion
The vector database query endpoint introduced in November now accepts a pre-computed embedding vector as an alternative to query text. It can also filter results by metadata, with exact matches on fields and matches on any value in an array field, such as tags. A metadata list mode returns every document that matches a filter without running a semantic search, with cursor pagination for large collections. A new endpoint deletes documents from a collection by document ID or metadata value, or clears the whole collection.
The chat and streaming inference APIs now handle conversations in which your application runs some of the tools. Set toolConfig.autoExecute to a list of tool names, and the platform runs only those tools. The API returns requests for the other tools to your application. Send their output back in toolResults to continue the conversation.
Improved Dashboard
GitHub connection with installation picker
Connecting GitHub, introduced in October, now asks you to sign in with GitHub first. The dashboard then lists the GitHub accounts and organisations where the Quant GitHub App is already installed. You can link an existing installation to your Quant organisation without installing the app again. If a linked installation was removed or lost access, the dashboard shows that the connection was lost so you can reconnect. The deployment wizards show the new connection status straight away, without a page reload.
Improved Dashboard
Cloud application deployments resume after a page refresh
If you refresh the page while a cloud application is deploying from a starter kit or a GitHub repository, the wizard now returns to the deployment's progress. It continues from its current step with your settings, instead of starting again.
Improved Dashboard
Deployment wizard early repository validation
When you deploy from a starter kit, the dashboard creates a GitHub repository from the template in your GitHub account or organisation. The deployment wizard now checks whether the target repository name already exists before it tries to create the repository. If the name exists, the cloud application starter kit wizard shows the error at the configuration step instead of at the end of the deployment process.
Improved Dashboard
GitHub template deploy reliability
When the wizard creates a repository from a template, it now retries automatically if GitHub returns a temporary error. The wizard also handles templates whose default branch is not main. It adds your application name as a GitHub Actions variable in the new repository, so builds deploy to the right application when its name differs from the repository name.
Improved SDKs
AI file storage and task classes
The TypeScript, PHP and Go clients are now at v4.9.0, generated from the unified API specification. They add classes for the new AI file storage and AI task endpoints, and the TypeScript client's crawlersRun method no longer requires a request body.
Starter kit deployments no longer wait for the first build
The last step of deploying a static site or cloud application from a starter kit could delay the deployment response by several minutes while the wizard waited for the first GitHub Actions build to finish. The wizard now completes as soon as the build starts and links to the GitHub Actions run so you can follow its progress.
Fixed Drupal AI provider
Image handling for vision requests
Vision and alt text generation requests did not receive image attachments correctly, so responses and generated alt text did not reflect the image. Attached images are now passed to these requests correctly.
Fixed Drupal AI provider
Tool calling and system prompts
In some request configurations, tool calling responses and system prompts did not work correctly. System prompts, such as an agent's instructions, were not applied. Tool calls were lost when the model requested more than one tool in a response. Both now work correctly.
Fixed Quant Cloud
Starter kit deployments with a managed database
Deploying a cloud application from a starter kit with a managed database could continue before the database was ready. The wizard now waits until the application and its database have finished provisioning before it continues. The managed database option is also now available for every template, and is selected by default when the template recommends a database.