Skip to content

Frequently asked questions

A platform for building AI agents and putting them on the channels your customers and teams already use, with every conversation kept in one auditable place. See How VirtuAI works.

You choose the provider and model per agent:

  • Google Gemini, through the Gemini API or through Vertex AI
  • Anthropic Claude, directly or through Vertex AI
  • OpenAI GPT
  • OpenCode Go, which offers models from other vendors such as DeepSeek, Qwen, GLM and Kimi

The agent form lists the models currently available for each provider. Models that the provider retires are removed from the list.

Yes. Each workspace adds its own keys in Settings: GOOGLE_API_KEY for Gemini, OPENAI_API_KEY for OpenAI, ANTHROPIC_API_KEY for Anthropic, and a Vertex AI service account and project for models on Vertex AI. See Workspace settings.

Can different agents use different models?

Section titled “Can different agents use different models?”

Yes. The model is set per agent. In the workspace web chat, an admin can also choose which models people are allowed to pick.

Web chat (as a link or embedded in your site), browser voice, WhatsApp through Twilio, Google Chat, Slack and Telegram. Agents can also be called by other agents over the A2A protocol, for example from Gemini Enterprise. Each agent has its own integration URLs, so different agents can serve different channels. See WhatsApp and Google Chat.

Agents answer in whatever language you instruct them to in their personality, within what the chosen model supports. The VirtuAI app and the web chat interface are available in English and Spanish. Gemini voice agents can be set to 22 languages and regional variants, including English, Spanish, Portuguese, French, German, Japanese, Chinese and Arabic.

Yes, through tools. You can connect your own HTTP APIs, attach predefined tools, or add MCP servers. Ready-made MCP servers include Gmail, Google Drive, Google Calendar, BigQuery, Cloud Storage, GitHub, Slack and HubSpot. Where a service needs the user’s own account, the user connects it once and the agent acts as that user. See Tools.

Can I test a change before my users see it?

Section titled “Can I test a change before my users see it?”

Yes. Turn on versioning for an agent to edit it in a draft, review the changes, evaluate the draft, and publish it as a numbered version. You can roll back to any earlier version. A workspace can also require someone else to approve a publish.

Use Evaluations: build a dataset of test cases, run it against an agent, and score each answer with automatic checks (for example, required text, expected tool calls, latency or cost) or with a model acting as a judge. Evaluations can also score a sample of real web chat conversations. Analytics and end-user thumbs up and down feedback complete the picture.

Yes. In Budgets, set a daily or monthly token or cost limit per provider or model. A budget either warns when the limit is exceeded, or blocks new requests until the period resets.

VirtuAI runs on Google Cloud. For the region that serves your workspace, contact support.

Your agreement with VirtuAI sets out data ownership. Within the product, everything you create belongs to your workspace, and nothing is shared between workspaces.

Yes. Agents, knowledge bases, tools, keys, conversations and users belong to a workspace, and access is checked on every request against the workspace you are working in.

One year by default. A workspace admin can set the conversations.retention_days setting anywhere from 1 day to 10 years (3,650 days). Older transcripts are removed automatically. See Conversations.

Yes. People with the conversations.manage permission can export the filtered conversation list as CSV, or a single conversation as Markdown or JSON. Exports mask sensitive data the same way the screen does, and every export is recorded in the access log. See Exporting.

How is sensitive data in conversations protected?

Section titled “How is sensitive data in conversations protected?”

Credit card numbers, API keys, private keys, passwords and bank accounts are detected as messages are stored, and shown masked. Revealing a value requires the conversations.pii_reveal permission, and every reveal is logged with the user and time. See Sensitive data.

With roles. Every workspace starts with admin, user and viewer roles, and each menu and action checks the permissions of your role. See Users and roles.

Yes. A workspace can sign in through its own OpenID Connect identity provider, with groups mapped to VirtuAI roles, and password sign-in can be turned off. Without SSO, people join by email invitation and sign in with a password.

How do I stop strangers from calling my agents’ webhooks?

Section titled “How do I stop strangers from calling my agents’ webhooks?”

VirtuAI verifies every channel with its platform’s own mechanism: Twilio’s signature for WhatsApp, Google’s token for Google Chat, the signing secret for Slack, and a webhook secret for Telegram. Requests that fail the check are rejected, so knowing an agent’s URL isn’t enough to talk to it. For voice connections, create an integration key and turn on enforcement. See Security overview and Integration keys.

How does billing work for Google Cloud Marketplace purchases?

Section titled “How does billing work for Google Cloud Marketplace purchases?”

If you bought VirtuAI through Google Cloud Marketplace, you are billed through Google Cloud Marketplace, on your Google Cloud billing account. Model usage with your own provider keys is billed by that provider.

Email admin@imvirtuai.com. Customers who bought through Google Cloud Marketplace get the same support as direct customers. See Support for what to include and how severity works.

In the release notes.