Skip to main content

Build, release, and monitor your first agent

Level: beginner

In this tutorial, you will build your first agent. Let's call it Scribe — a business-writing assistant that turns a colleague's rough notes into a polished email, message, or summary.

Writing is a deliberate choice for a first agent. It is something a capable model already does well, so Scribe needs no knowledge sources, no tools, and no skills of your own: a prompt alone makes it useful. That means you can learn the platform end to end without first waiting on data access, a content review, or a governance sign-off.

You will then publish Scribe, deploy it to an environment, and read the trace of a real run. By the end, you will have taken one agent through the full lifecycle — build, test, release, observe — and you will know where each of those stages lives in Creatio AI Studio.

Before you start​

  • Access to Creatio AI Studio with permission to create an agent and publish it.
  • A configured model provider. Without one, agents cannot make LLM calls. Learn more: Connect a model provider.
  • A Sandbox environment to deploy into. Learn more: Set up an environment.
  • If you work in a shared workspace, prefix the agent name with your initials so it does not collide with a colleague's.

Stage 1. Create the agent​

  1. Go to the Agents section in the navigation panel.
  2. Click New → Personal agent.
  3. Select Prompt Agent Designer and click Next.
  4. Enter the name Scribe, prefixed with your initials.
  5. Enter the description Business-writing assistant: drafts, rewrites and polishes emails and messages.
  6. Leave the Code field as Creatio AI Studio generated it. A personal agent receives its final code from the server, so anything you type here is replaced.
  7. Click Create.

The agent opens on the Agent tab of the Build tab group, with an empty system prompt. It arrives bound to the shared Creatio.ai Twin channel and carries the vendor skills every agent gets, such as Creatio date and time handling. You add nothing to it in this tutorial.

Stage 2. Give the agent its instructions​

The system prompt is the whole brain of this agent. Everything Scribe does comes from these instructions.

  1. Paste the following into the System prompt field.

    You are Scribe, a business-writing assistant. You help people write and
    improve everyday work communication: emails, chat messages, meeting notes,
    short announcements, and summaries.

    What you can do:
    - Draft a message from a few rough notes.
    - Rewrite text to be clearer, shorter, or more polished.
    - Change the tone (for example, more formal, friendlier, more direct).
    - Summarize long text into key points or action items.
    - Fix grammar and structure without changing the meaning.

    How to respond:
    - If the request is clear, just produce the writing - do not over-explain.
    - If key details are missing (audience, goal, tone, length), ask one short
      question first, then write.
    - Keep business tone by default: clear, respectful, concise.

    Stay focused on writing help. Politely decline unrelated requests.

  2. Enter the Welcome title Scribe - your writing assistant.

  3. Enter the Welcome Message Hi! Paste any rough notes or text and tell me what you need.

  4. Check the model in the Model field. A new agent starts on GPT-5.6 Terra, a Creatio-provided model, or on the Default model that your organization selected. Keep it for this tutorial.

  5. Leave Temperature as it is. GPT-5.6 Terra accepts only a fixed temperature of 1, so the slider is locked. With a model that accepts a custom temperature, the slider runs from 0 to 2 and starts at 0.7.

  6. Confirm that Memory is turned on. It is on out of the box.

The editor saves on its own as you type, so there is nothing to click. The header reports "Saving…," then "Saved," and the Agent tab shows your prompt and the selected model.

Stage 3. Test the agent​

  1. Go to the Preview panel on the right. It opens in Text mode, which is what you want here.

  2. Select Sandbox in the Environment field.

  3. Check that Show trace is on. It is on out of the box, and it is what lets you follow the run in Stage 6.

  4. Send the following message.

    Turn these notes into a polite email: can't join Thursday call,
    send recap, will review by Monday.

  5. Send a second message to check the agent's boundary.

    Write me a short poem about the ocean.

Scribe returns a usable email for the first message, and politely declines the second — the refusal rule in your prompt is working.

Stage 4. Publish a version​

Publishing freezes your draft into an immutable version. You never edit a released version — you supersede it with a new one, which is what makes a rollback trustworthy later.

Click Publish in the top right. The header now reads "v1 · Current" and the agent becomes read-only: a published version is a fixed record of what you released. To carry on editing, click Create draft, which opens a new draft on top of v1.

Stage 5. Deploy to Sandbox​

  1. Click Deploy in the top right.
  2. Select Sandbox. The row reads "Deploy v1."

The deployment starts as soon as you select the environment, with no confirmation step, and Creatio AI Studio reports "Deploying v1 → Sandbox…." When it finishes, Sandbox shows a Delivery of "In sync" and a Runtime of "Ready," or "Idle (on-demand)" while nothing is calling the agent.

Stage 6. Read a run trace​

Every execution leaves a trace. Reading one now means you know where to look when an agent later behaves oddly.

  1. Go to the Observability section under Runtime in the navigation panel.
  2. Find your run in the list. Filter by Environment if the list is busy.
  3. Open the run.
  4. Review the General run information section: last activity, initiator, environment, agent version, and the start and completion times. The run status is a tile above this section, next to duration and the call counts.
  5. Click a span in the Execution timeline.
  6. Read its Input and Output.

You can see the exact text sent to the model and the exact reply it returned — not an inference from the answer, but the call itself.

As a result, you have an agent that is built, tested, published, deployed, and observable. You have seen the four surfaces every future agent uses: the designer, live preview, the deployment controls, and the run trace.

Next steps​

Scribe works from a prompt alone. You can enrich its capabilities by adding the following functionalities.

See also​

Choose a use case and define success, boundaries, and required data

Create a prompt agent

Monitor agent runs and read traces