Create a prompt agent

This article explains how to create a new prompt agent in Creatio AI Studio step by step. In general, prompt agents are the easiest to create. You only need to describe what the agent needs to do in plain text. You can also enhance the agent with additional tools, skills, and knowledge sources as well as specify its triggers, version the agent, and deploy it to different environments. After the agent is created, you can observe its behavior using a full-fledged analytics suite and govern the agent via dedicated policies.

Example

Create an FAQ agent that answers common questions based on the knowledge library in the web chat. Users must be able to upload files.

Log in to Creatio AI Studio.
Go to Managed Agents group → Agents. This opens a page.
Click New → select the required agent type. Select "Personal agent" if you only need an agent for personal use, select "Enterprise agent" if the agent must be visible and usable across your organization.
Select Prompt Agent Designer. This opens the "Name your agent" step.

Fill out the general agent identity fields.

Field	Field value
Name	Human-readable name for the agent. Use short, clear names that reflect the agent's main goal. For example, "FAQ Agent."
Code	Technical ID used by Creatio AI Studio and the LLM. Creatio AI Studio generates this value automatically as you type the name, for example, "builder-faq-agent-c0f758." You can edit the code of an enterprise agent. A personal agent receives its code from the server, so a code you type there is replaced.
Description (optional)	Plain-text summary of what the agent does and when to use it. For example, "Answers the common questions about the company's products."

Click Next.
Bind channels to the agent on the Assign channels step. Use the Create new tab to add a channel, or Reuse existing to pick one that is already registered. For this example, bind the "Creatio.ai Twin" channel, whose card reads "Shared Twin channel will be reused." You can skip this step and bind channels later, but users cannot reach the agent until at least one channel is bound.
Click Create. This opens a page on the Build → Agent tab.

A brand-new enterprise agent has nothing published yet, so the version, Publish, and Deploy controls appear once you make your first edit.

Fill out the System Instructions block.

Field	Field value
System prompt	Internal instructions sent to the model at runtime. Use this field to define the agent's role, behavior, and constraints. For example, "You are an FAQ agent who helps users find answers to questions about the product. Use only information from the connected knowledge base. Respond briefly, clearly, and in the user’s language. If there is no exact answer, do not make one up. Indicate that no information was found and suggest that the user contact support. Whenever possible, include a link to the source."
Welcome title	Heading displayed when a chat conversation starts. For example, "Do you have any questions about the company knowledge?"
Welcome Message	Opening greeting used when a chat or voice conversation starts. For example, "Hello! How can I help you today?"
Model	The large language model (LLM) the agent uses. A new agent starts on GPT-5.6 Terra, a Creatio-provided model, unless an administrator selects another model in Default model on the Platform LLM page. With "Use platform default," the agent follows that model, also when an administrator changes it later. Models that Creatio provides are listed under Creatio provided, and a model your organization serves with its own key is listed under its vendor. Learn more: Connect a model provider.
Temperature	Slider that controls response randomness on a scale from 0 to 2. A higher value produces more varied responses; a lower value produces more focused, deterministic ones. Set to "0.7" out of the box. On a model that accepts only one temperature, such as GPT-5.6 Terra, the slider is locked at that value, "1.0."
Memory	Toggle to enable session-level memory so the agent can retain context across turns in a conversation. Enabled by default. Expand Memory settings to clear what the agent has remembered.

Go to Build → Tools tab to select which built-in tools the agent is allowed to call.
1. Review the tool groups: Built-in model tools, Generative UI tools, Activity tools, Built-in tools, and Other tools. The chips above them, such as "Available" and "Enabled," count the tools in each state. They are counters, not filters.
2. Enable a tool by selecting the checkbox next to its name, and clear a checkbox to disable one. A new agent starts only with the built-in tools its model supports, such as Web search. For this example, clear Web search, so the agent answers from your knowledge library rather than the open web.
  
  Built-in model tools run inside the model itself, so each one is available only when the selected model supports it.
3. Click Add HTTP tool to register a new external tool. Learn more: Add an HTTP tool.
Go to Build → Integrations tab to enable the tools that external systems expose.

HTTP and MCP tools are enabled here, one by one, rather than on the Tools tab. Each integration lists its tools with an Enabled checkbox and starts with none of them enabled. Choose an Access mode for the integration, then select the tools this agent may call. Learn more: Integrations in Creatio AI Studio.
Go to Build → Skills tab to attach skills to the agent.
1. Select an existing skill in the Attach skill picker, which searches published skills, then click Attach. Once attached, the skill's files, scripts, references, and assets are displayed in the panel. For this example, do not attach any skills.
2. Verify that the Enabled toggle is turned on for the attachment. A skill that is attached but disabled never runs.
3. Reference the skill in the system prompt so the agent knows when to invoke it. The prompt says when; the skill says how.
  
  A skill takes effect only after it is published. Editing a skill changes its draft, and the draft does nothing until you publish a new version of it. Learn more: Create a skill.
  
  When an environment serves an older release than the latest one you published, the Skills tab shows "Deployment update required" and names each such environment. Deploy or promote the latest published version to those environments. The notice clears once every active environment serves it.
Go to Build → Knowledge tab to attach knowledge sources the agent can query for context.
1. Open the Select a knowledge source... list.
2. Select a knowledge source. The list attaches one source at a time, so repeat the step for each source you need. For this example, select your knowledge base.
3. Review the Citations toggle on each attached source. It is on out of the box, so the agent names the source it answered from.
Once attached, the agent uses the indexed content from the selected knowledge sources to generate more accurate and context-aware responses.

Go to Configure → Profile tab to set how the agent is presented to users in chat.

Field	Field value
Description	Short explanation shown in agent cards and compact previews.
Full Description	Longer agent description, operating mode, and guidance for users.
Avatar style	The icon and color shown on the agent's card. Reset to default avatar restores the generated one.

Everything on this tab is customer-facing: it is what users see, not how the agent behaves.

Go to Configure → Channels tab to bind channels to the agent.
1. Click Add Channel. Use the Create new tab to add a channel, or Reuse existing to link a channel that is already registered in Creatio. For this example, select the "Web chat" channel, then click Add Channel in the dialog.
2. Review the channel summary widgets: Bound channels, Active, and Voice.
The agent is not reachable by end users until at least one channel is bound.
Go to the Preview panel on the right to test the agent in live preview.
1. Select the target environment, for example, "Sandbox," in the Environment field.
2. Type a message in the Send a message… area and press Enter. The panel starts in the Text mode, which shows the replies only, not how a channel delivers them. Once channels are bound in the selected environment, Preview as also offers a mode for each bound channel type, such as Web chat or Creatio.ai Twin.
  
  To dictate the message instead, click Dictate with microphone and speak. When you finish, click Stop and insert text to review the text in the message box. To send it right away, click Send now. To cancel, click Discard recording.
3. Review the model's response and the reasoning trace below it. Show trace is on out of the box. When the reply is finished, every step of it is shown as completed.
4. Read the trace to confirm the agent used the resources you attached, rather than guessing. The trace shows the following.
  - Retrieved knowledge — which source answered, and the chunk that fed the model. No retrieval on a question your knowledge covers means the source is not attached, is still indexing, or has a description too vague for the agent to pick it.
  - Skill loads — which skill fired. A skill that never loads is unattached, disabled, or unpublished.
  - Tool calls — which tools ran and with what arguments.
  - Policy events — where a governance policy transformed the payload. A PII mask appears here as the placeholder rather than the raw value. Learn more: Review policy decisions.
Click Publish in the top right.
Click Deploy in the top right and select the target environment.

Neither action asks you to confirm: the version is published, and the deployment starts, the moment you click.

As a result, Creatio AI Studio publishes the agent and the published version becomes the current one. The agent becomes available to users of the environment you deployed to, through the assigned channels.

Configure agent runtime limits

This step is optional. Runtime limits cap what a single run of the agent may consume, so no one run can exhaust the budget or hang.

Open the agent.
Go to Configure → ⋯ → Advanced tab. The tab sits in the overflow menu of the Configure group, next to Deployments.

Fill out the limit fields.

Field	Field value
Task timeout queue (seconds)	Maximum time a task may wait in the queue to be picked up by a runtime before it fails. Separate from Run time budget (seconds), which limits how long the run executes once it starts. Set to "900" out of the box; the maximum is 86400 seconds, which is 24 hours. Listed in the Resources section.
Run time budget (seconds)	How long one run may execute, whichever channel, test chat, or API call starts it. Accepts a value from "10" to "3600," and is set to "600" out of the box. Listed in the Runtime limits section.
Max turns	Maximum model turns the runtime may execute before it stops a run. Accepts a value from "1" to "300," and is set to "200" out of the box. An agent with more than 5 tools needs at least 15 turns, otherwise the save fails. Listed in the Runtime limits section.

As a result, Creatio AI Studio caps every run of this agent at the configured limits. Channels and the AI Twin chat use the new limits once you publish and deploy the version that carries them, and the AI Twin chat waits for an answer for as long as the run time budget allows, up to 3,600 seconds. These are per-agent guardrails and sit alongside the organization-wide budget. Learn more: Set an AI budget.

Configure agent runtime limits​

See also​

Configure agent runtime limits

See also