Bot & Flows

In the Bot Flow builder, wire an AI Agent node to a Chat Model and a Vector Database node backed by the RAG knowledge base, which ingests content via URL crawl, sitemap, file upload, or Google Drive. A Guardrails node keeps responses on-policy so the agent answers strictly from your sourced material rather than improvising. Re-crawl or re-upload to refresh the knowledge base whenever docs change.

Drop an If-Condition node into the flow to read the customer's reply or a captured variable and split the path accordingly, sending each branch to its own AI Agent, message, or downstream node. Because branching lives on the same canvas, designers can see and adjust every route visually instead of editing code.

Insert a User-Approval node at the decision point so the flow pauses and routes the proposed action to a human for review before it executes. Combine it with an If-Condition node so approval is only required above a chosen threshold, keeping low-risk paths fully automated. The rest of the flow continues automatically once a person approves or rejects.

Use the Keyword Flow builder to launch a WhatsApp form the moment a customer sends a chosen keyword, collecting fields conversationally inside the chat. A resume-if-abandoned prompt re-engages people who started but didn't finish, so partial sessions are recovered rather than discarded.

Build the conversation once in Bot Flow and attach multiple Trigger nodes to bind it to WhatsApp, Instagram, and Telegram (and other supported channels) from the same canvas. The channel binding spans WhatsApp, Telegram, Instagram, Discord, Facebook, SMS, Email, RCS, and a generic webhook, so one source of truth serves them all.

Use the Call Flow builder to design an inbound IVR visually, with a node to detect voicemail, an option to transfer callers into a queue, and a path to capture a voicemail when no one can take the call. The WhatsApp Call and HTTP nodes are available on the same canvas to extend the voice experience further.

Use the built-in chat and voice simulator to run the flow as a chosen persona and language before publishing, exercising every branch safely. Trace logs show the path the bot took step by step, and you can copy them for sharing or debugging, so issues are caught and documented prior to go-live.

Drop an HTTP Request node into the flow and import an existing cURL command to configure the call in seconds, then feed the response into an If-Condition node to branch on what the API returns. For richer integrations, MCP and Web Search nodes are available on the same canvas to bring external tools and live results into the conversation.