Cloud AssistantCopy Anchor Link

Select a project in Console, then select the Cloud Assistant sparkle icon immediately to the left of notifications in the top bar. You can also select Open Cloud Assistant in the fixed deployment card on project Home. The assistant opens in a right-side panel without leaving your current screen. Enter a question and select Send. The icon is unavailable until a project is ready and you have access to its metadata. The assistant checks the resources you have permission to read before answering.

The question field grows automatically up to eight lines, then scrolls internally. Manual resizing is disabled. While the question field is focused, press ⌘+Enter on macOS or Ctrl+Enter on Windows/Linux to send. Enter inserts a new line.

Try asking:

• “Which containers are in this project?”
• “List my PostgreSQL and Redis databases.”
• “What CPU and memory are assigned to my containers?”
• “Which resources are still provisioning?”

What it can readCopy Anchor Link

The assistant reads resource names, resource IDs, stored lifecycle states, update times and assigned plan specifications for Container Engine, Memory Database and Relational Database. It returns at most 50 resources per authorized product and identifies partial lists. Open the product list for the complete inventory.

Lifecycle state is not live health: Active does not prove an application is responding. The assistant does not inspect live metrics, logs, environment variables, credentials or secret values. Resource queries do not change anything. The reviewed flows below can create new HTTP containers, Redis and PostgreSQL databases, and propose lifecycle actions for an existing Container Engine resource. Configuration editing, commands and database lifecycle actions are not supported.

Project access and each product's existing read permissions apply. Products you cannot read are excluded; their omission does not mean the project has no such resources.

The assistant is designed to decline unrelated questions (such as store locations) and requests outside its supported capabilities. Keep each question focused on your selected project; mixed requests containing unrelated questions are also declined. Replies follow your question's language.

Project costsCopy Anchor Link

Ask “What is this project's estimated monthly cost?” or “How much usage has been recorded this month?” Billing questions require workspace billing-history access as well as access to the project and relevant products.

The assistant shows per-resource hourly rates and a monthly estimate based on 730 hours. Existing billing rates take precedence over current catalog prices. Estimates include allocated resources even when stopped; they are not a prediction of future uptime or a final invoice. Unknown prices and partial resource lists are identified rather than treated as free. Estimates use USD without inventing currency conversions, taxes or discounts.

Current-month usage is shown separately: recorded charges for usage intervals ending in the current UTC month, through the displayed observation time. It includes deleted resources in products you can read, can lag usage that has not been processed, and does not indicate whether a payment was made. The assistant does not disclose other projects, payment methods or account balances.

Operate an existing containerCopy Anchor Link

Ask “Restart web” or “Stop the container with ID …”. The assistant can propose Start, Stop, Restart, Redeploy or Delete for one Container Engine resource in the selected project. Redeploy requires an application resource. If names are ambiguous, supply the resource ID; the assistant does not choose an arbitrary match. The candidate list is limited to 50 resources. For a resource outside that list, use its detail page.

The card displays the project, resource name, resource ID, action and service impact. Select Review action, then confirm the action in the dialog. For Delete, enter the exact displayed resource name. Sending a chat message, including “yes”, never executes the action. Proposals expire after 15 minutes and a new question supersedes previous unsubmitted cards. Container read permission and the corresponding start, stop, restart, deploy or delete permission are required. Your permissions must still allow the action when you confirm it.

Request accepted means the action was submitted and may still be in progress. Use View resources to inspect progress in the resource screen. Redeploy applies saved configuration and Envault values; mutable image tags can resolve to a newer image. Stopping does not release allocated resources or end their charges. Delete follows the existing deletion workflow, including its retention behavior.

Each card sends its action at most once. If confirmation is interrupted or access changes after submission, the outcome may be unconfirmed. Check the resource before requesting another action; the card never automatically resends it. Closing the panel or changing project does not undo an action already accepted by the server.

Deploy an imageCopy Anchor Link

Ask something like “Deploy nginx:alpine on the smallest available plan” or “Deploy ghcr.io/my-team/my-app:1.0; it listens on HTTP port 3000.” For an unfamiliar image, the assistant asks for missing runtime information. It cannot verify an image's requirements from its address alone.

Review the proposal card: image and tag, name, server, internal HTTP port, plan CPU/RAM, hourly price and a monthly estimate based on 730 hours. Estimates cover the plan only. Other usage and charges may differ. The image's default command and production environment are used; existing Workspace defaults apply. Proposals expire after 15 minutes. Ask for a revised proposal if you want different settings.

Optionally select an existing Secret Group and environment under Link Envault in the card. Its values are supplied during deployment, without being sent to the assistant. Linking requires container binding and workspace secret access permissions. With no selection, existing Workspace defaults still apply. Create any new secrets in Envault first; never enter them in chat.

Select Provision to authorize the proposal and your optional Envault selection. No resource is created just by sending a message. Existing project creation permissions, billing and available credits apply. This flow also requires container and domain read permissions so it can show progress. Changed settings or plan pricing require a fresh proposal.

The card checks progress every five seconds. Successful container creation is followed by the generated domain's HTTPS check. Deployment ready · HTTPS endpoint connected means the background readiness check completed, not a guarantee of every application function or ongoing health. Use View container to inspect the resource, logs and domains.

If deployment confirmation is interrupted, use Check again to resume status checks for the existing deployment. A failed or unknown deployment never triggers automatic recreation. Check the container list before requesting another deployment. Checks pause after 15 minutes; you can resume them. Closing chat or changing project stops local checks but does not cancel an accepted deployment; the resource list and notifications retain its progress.

This first version supports single HTTP applications. Configure inline environment values, volumes, custom commands and private-registry credentials through the regular Container Engine creation workflow instead. Never paste secrets into chat.

Create Redis or PostgreSQLCopy Anchor Link

Ask “Create a Redis database on the smallest available plan” or “Create a PostgreSQL 17 database.” The assistant selects from available versions and compatible server plans and asks for clarification when needed. Review the database version, server, plan CPU/RAM, storage allowance and estimated price before selecting Provision.

Provisioning requires the corresponding product's create and read permissions, a billing account and sufficient credits. Credentials are generated by the platform and are not displayed in chat. PostgreSQL uses the app database. This flow does not automatically export database credentials to Envault or connect the database to an application.

The card checks the existing provisioning operation every five seconds. Database provisioning complete means RDC has confirmed creation; it is not a test of access from your application's network or the internet. Select View database for connection details and access settings. A missing or interrupted creation response never automatically creates a second database; Check again only resumes status checks.

Response displayCopy Anchor Link

The assistant shows progress while checking your request and reading resources. Answer text appears as it is generated. Closing and reopening the panel preserves the current answer; New chat cancels the active request.

Conversations and privacyCopy Anchor Link

Your messages and authorized resource metadata, including cost information when you ask billing questions, are sent to Google Gemini to generate an answer. Do not enter passwords, API keys or other secrets. Answers can be inaccurate; verify important details in the resource screen.

Conversation history is kept only in the current page. Closing and reopening the panel on that page preserves its conversation and unfinished question. Close it with the close button or Escape; focus returns to the Assistant icon. New chat, leaving the page, refreshing it, or switching accounts or projects clears the conversation. A permission change also resets it. RDC does not save conversation history. Google Gemini processing is subject to the provider's data policies, which vary by API tier. Free-tier messages and responses may be used for product improvement and reviewed by humans. RDC's in-memory chat does not imply zero retention by Google.

Each question accepts up to 4,000 characters. Conversations are limited to five question-and-answer turns and a bounded total length. Start New chat when the limit is reached.

Unavailable answersCopy Anchor Link

If a request fails or its stream is interrupted, select Retry to send the same question again. Retrying replaces any incomplete answer without duplicating your question. Requests are rate limited; wait briefly before retrying. If the assistant remains unavailable, continue using the product screens and contact support with the displayed error. A failed chat request does not change your resources.

The Cloud Assistant icon shows a blue activity ring while a reply is being prepared or streamed, even when the panel is closed. The ring stops when the reply completes or fails. Resource provisioning progress remains available in Notifications.