Skip to main content
The April 2026 PolyAI Agent Studio release introduces first-class multi-language support – configure multilingual agents entirely from the UI, manage translations in a dedicated page, connect external tools with MCP integrations, and script agent configuration through the new Agents API. Expand the items below for details:
Build a single agent that handles multiple languages – no separate projects, no manual start_function workarounds. The new Multi-language settings page lets you configure everything from the UI.What’s new:
  • Add languages in the UI – go to Behavior and add up to 10 additional languages with a couple of clicks.
  • Per-language voices – assign a dedicated Agent voice and Disclaimer voice for each language under Voice > Settings, with multi-voice support per language. The agent switches voice automatically when the conversation language changes.
  • Test in any language – a new language dropdown in Agent Chat lets you start conversations in a specific language to verify behavior before going live.
  • Conditional content tags – wrap language-specific content in <language:en> / <language:es> tags to serve the right version from a single prompt.
  • Test in any language – a new language dropdown in Agent Chat lets you start conversations in a specific language to verify behavior before going live.
  • Language metadata – see which language was used for each conversation in Conversation Review (list and detail), and manage language-tagged audio in Audio Management.
See Multi-language for full setup instructions.
Fine-tune how your agent speaks in every language. The new Translations page under Voice > Advanced gives you a central place to manage auto-translated content and override anything that doesn’t sound right.What’s new:
  • Translation cards – each card holds a piece of content (like a greeting or confirmation message) with versions for every configured language.
  • Automatic translation – when you save a card, all configured languages are translated instantly. Cards that haven’t been manually reviewed are marked as “Auto Translated”.
  • Manual overrides – edit any translation to fix awkward phrasing, adjust formality, or handle idioms that don’t carry over. Overridden entries are marked as “Manually Translated”.
  • Insert anywhere – reference translation cards via the action menu in greetings, disclaimers, behavior rules, prompts, delay control responses, and SMS templates.
  • Access in functions – use conv.translations.your_key to pull the right translation programmatically for hard-coded utterances.
See Translations for setup instructions.
Script agent configuration and deployment through a new public REST API. The Agents API covers the same surface as Agent Studio’s build pipeline — create agents, edit behavior, manage the knowledge base, configure variants, and promote deployments — all from code.What’s new:
  • Full CRUD for agent configuration – create, update, duplicate, and delete agents; read and update behavior rules on any branch.
  • Knowledge base management – list, create, update, and delete knowledge base topics programmatically, making bulk migrations and CMS sync straightforward.
  • Variants and attributes – sync multi-site variants from a source of truth (locations database, CRM) instead of editing them by hand.
  • Deployment pipeline – publish drafts to Sandbox, promote through Pre-release to Live, and roll back from CI.
  • Telephony plumbing – import phone numbers, reassign them to different connectors, and look up which connector serves a given number.
  • Branches and real-time configs – manage branches and real-time configuration values alongside the rest of the agent.
See the Agents API reference for the full endpoint list.
Agent Studio now supports Model Context Protocol (MCP) integrations, letting you connect external tool servers to your agent directly from the UI.What’s new:
  • A dedicated MCP tab is now available under Integrations.
  • Add an MCP server by providing a URL, authentication details (Header, Query parameter, or OAuth), and an optional timeout.
  • Agent Studio auto-discovers the tools available on the connected server and displays them in a configuration panel.
  • Toggle individual tools on or off to control which functions are exposed to the agent, helping manage context window usage.
  • Supports Header, Query parameter, and OAuth authentication methods using secrets from the Secrets Vault.
Projects that previously configured MCP through experimental config will continue to work – the old MCP tab is hidden once the feature is enabled, but existing servers keep running in the background. The old config only stops being used if you add a new MCP server through the new UI. See MCP integrations for setup instructions.
Last modified on July 7, 2026