AgentBuilder

AgentBuilder is for owners who want more than a chatbot. A chatbot answers. An agent acts. It reads the knowledge base before replying. It pulls a product from the catalog and drops the card in the chat. It writes a new lead into the CRM. It walks a customer through a scripted flow step by step.

Every ability the agent can use is a simple toggle in the editor. Knowledge base for domain expertise. Business profile for "who we are" context. Catalog for products and pricing. CRM for customer history and writing new records. Scenario for guided flows. Turn them on, and the agent has the capability. Turn them off, and it stays out.

Creating an agent takes three inputs: a system prompt (picked from the prompt library you already built), a name and description, and a model chosen from the platform's approved list. The prompt is cloned into the agent so you can iterate on the agent's version without breaking your master prompt.

Descriptions are generated automatically in the background — the owner does not have to sit and write marketing copy about their own agent. Models are curated: the admin-approved list prevents users from picking something too weak or too expensive for the role.

Context size is managed deliberately. The agent does not dump every knowledge-base document into every prompt — it pulls only what is relevant, keeping responses fast and on-topic. The same discipline applies to CRM and catalog lookups.

A hub lists every agent in the workspace, with names, descriptions, and the tools they have enabled. Admins can view and edit any agent across the platform for support. Each agent is independent — tuning one never affects another.

Once configured, the agent attaches to a chat from ChatBuilder and starts handling real conversations. If an agent underperforms, you tweak it or swap it; no history is lost. A sales agent, a support agent, and an onboarding agent can all coexist in one workspace, each specialised for its lane. This is how the platform scales past "one chatbot per site".