Create an agent with AI Twin
AI Twin builds agents from a plain-language description. Instead of filling out the Prompt Agent Designer field by field, you describe what the agent should do, and AI Twin composes it — identity, system prompt, and, when you ask for them, skills, knowledge sources, and integrations.
AI Twin is the chat on the Home page of Creatio AI Studio. It is the landing surface out of the box. Learn more: Work with AI Twin.
What AI Twin can do
AI Twin works through a set of capability bundles. Each one covers a stage of the agent lifecycle. While AI Twin is the selected agent, the Commands and skills menu in the message box lists them. Type / in the message box to open the menu.
Capability | What it does |
|---|---|
Agent Composer | Creates an agent from your description, and attaches skills and integration tools to it. |
Agent Refiner | Changes an existing agent — its configuration, its attached skills, and its integration tools. |
Agent Inspector | Reads back an agent's configuration, versions, environment bindings, attached skills, and run traces. |
Deployment Manager | Publishes a version, deploys it to an environment, and redeploys or rolls one back. |
Knowledge | Creates and manages knowledge sources. |
Host Navigation | Opens the matching Creatio AI Studio page for whatever you are discussing. |
Scheduled Tasks | Prepares a scheduled trigger from a plain-language description, for you to review and confirm. Scheduled triggers are enabled per organization and appear only where they have been turned on. |
Create an agent
-
Go to the Home page in the navigation panel.
-
Describe the agent you need in the chat, in plain language. Cover what it does, who it is for, and how it should respond. Say whether it needs knowledge sources, tools, or skills.
-
Review the agent preview card AI Twin returns, then open it to read the full summary in the side panel.
The card also offers Deploy while the agent has no deployed version or has changes that are not deployed yet, and Use once the agent is ready, which adds it to your agents in the chat. If a deployment fails, the card reads "Deployment needs attention."
Element
What it shows
Status
Where the agent stands: "Created," "Draft," or "Published."
Readiness
Whether the agent is usable as composed: "Ready," "In progress," or "Needs attention."
System prompt
The instructions AI Twin wrote. Read these before you rely on the agent — they are the whole of its behavior. They also tell the agent, when a tool call fails or data cannot be read, to say in plain language what failed and what it means, so keep that instruction when you edit the prompt.
Skills, Knowledge, Integrations, Channels
What AI Twin attached. Each section reports when nothing is attached yet.
-
Refine the agent by replying in the same conversation. Ask AI Twin to change the tone, tighten the prompt, attach a skill, or add a knowledge source. You can also move to another task in the same conversation, for example, working with Creatio records, and return to the agent later.
-
Click Open full agent editor at the bottom of the side panel to continue in the Prompt Agent Designer.
As a result, Creatio AI Studio saves the agent as a draft. If you also ask to make the agent live, for example, "create it and deploy it," AI Twin publishes and deploys it right away. Otherwise, AI Twin offers to deploy it and deploys it only when you accept, for example, when you reply "yes." AI Twin deploys the agent to your current environment or to the environment you name. You can then edit, test, and publish the agent like any other. Learn more: Create a prompt agent.
A description that produces a working agent in one turn.
Create a simple personal agent called Jargon Buster. When I give it any business
or technical term or acronym, it explains what it means in one or two plain-English
sentences and adds a quick everyday example. Keep it friendly and concise.
It doesn't need any knowledge sources, tools, or skills.
Test the agent
- Open the agent in the Prompt Agent Designer.
- Go to the Preview panel on the right.
- Select the environment in the Environment field.
- Send a message that exercises the agent's main job.
- Read the trace below the answer to see how the agent reached it. Show trace is on out of the box.
As a result, you can confirm the composed prompt behaves the way the description intended before you publish the agent.
Read the system prompt AI Twin wrote before you publish the agent. AI Twin composes from your description, so anything you left unsaid — tone limits, topics to decline, how to handle missing information — is AI Twin's interpretation rather than your decision.