Bring the same AI system to every supported channel

A Chatbot is the AI composition. An Integration is its connection to an external channel. Each Integration keeps its own provider, authentication, presentation, limits, and operational status without moving channel concerns into an Agent.

The setup path follows the provider instead of forcing every channel through one authentication model.

For an OAuth connection, authorize the provider, return to the connection flow, review the available remote channels, choose the Page or Official Account, and then configure the Integration. The product manages connection-session details internally rather than presenting them as public Integration identity.

Select the configured Chatbot, give the Integration a display name, and add its description, opening message, and rate limit. Non-Web channels also expose blocked user IDs. The external channel plus the configured Chatbot becomes the Integration relationship.

Display name, description, opening message, presentation fields where applicable, and status belong to the Integration. Agents retain focused responsibilities, while the Chatbot retains the reusable AI composition.

A Web Integration can carry a display name, avatar, primary color, greeting message, opening message, description, allowed domains, rate limit, and status. These are explicit fields in the current workspace, not a claim of a separate visual widget editor.

The Web create response includes plain_widget_public_key and embed_script. The completion view exposes those generated values so the configured widget can be placed on a website. This page never includes a real customer key.

Use the allowed_domains Integration setting to list the website domains intended to host the widget. It is a focused deployment control; this page does not turn it into a broader compliance or security guarantee.

Start the Messenger OAuth flow, return to the available Page list, select the intended Page, complete Chatbot Setup, and create the Integration. The selected Chatbot remains the AI system behind the Messenger connection.

Enter the Telegram bot token, validate it, confirm the bot identity returned by the connection, select the Chatbot, and complete the Integration. This describes the current connection flow without claiming payments, groups, commands, or inline mode.

Zalo OA uses OAuth and external channel selection. Zalo Bot uses token validation. The Integration model preserves that distinction with channel_type = oa or bot, so the two connection types do not collapse into one flow.

Non-Web Integrations currently support blocked_user_ids as Integration-level configuration. It provides a channel-specific list to manage; it is not presented as universal abuse prevention.

The rate_limit_per_minute setting keeps the configured limit attached to one Integration. Set it for that connection's expected use rather than assuming unlimited throughput.

The current Integration status model exposes active, disabled, and error. Where management controls are available, disabling preserves the Integration configuration while stopping its active use; enabling makes the retained connection active again.

Integration detail can show the connected Chatbot, platform, status, created and updated timestamps, and last event or last active information where available. These fields support practical inspection without claiming full observability.

The current workspace exposes Human Handoff only for Messenger and Zalo OA. It is not exposed for Web Widget, Telegram Bot, or Zalo Bot. On those supported Integrations, settings include enable or disable, handoff conditions, pending timeout, and idle timeout.

Agent configuration, documents and data, Skills, MCP tools, and Memory belong to the AI system rather than each channel Integration. Keeping that composition centralized avoids maintaining a separate intelligence stack for each interaction surface, without promising instantaneous propagation or zero downtime.

Opening message, rate limit, blocked users, Web presentation, and status are channel or Integration concerns. Update those settings for the connection without turning them into Agent instructions.

An Agent provides a capability. A Chatbot composes capabilities and Memory. An Integration defines where interaction happens. Automation defines when scheduled work runs, and a Run records one scheduled execution.

Connect the supported interaction surfaces relevant to the use case. More channels are not automatically better; each one adds provider authentication, configuration, presentation, and operational state that should have a clear purpose.