> For the complete documentation index, see [llms.txt](https://docs.autommate.com/userguide/llms.txt). Markdown versions of documentation pages are available by appending `.md` to page URLs; this page is available as [Markdown](https://docs.autommate.com/userguide/ai-agent-composer/configure-an-agent-in-ai-agent-composer.md). # Configure an agent in AI Agent Composer AI Agent Composer brings an agent's model, instructions, knowledge, skills, channels, and conversation review into one workspace. Start with one clear purpose, then add the information and capabilities required for that purpose. AI Agent Composer 1.5.3 has six tabs: Agent Details, Knowledge, Skills, Channels, Playground and Logs. Configure the model and instructions in Agent Details. Use the Agent Details reference below to find the controls, then follow the onboarding-policy walkthrough at the end of the page. It takes you from a saved configuration to a small text source and tested replies.
Agent Details tab and model configuration in AI Agent Composer

Use Agent Details to choose the model connection and set the agent’s purpose and instructions.

### Decide what the agent should do Write a one-sentence purpose before configuring the model. For example: > Help employees find answers in the approved onboarding policy and explain when a question needs a human owner. This example is deliberately narrow. A knowledge-answering agent does not need the same skills or channel access as an agent that changes records in an application. ### Configure the model Open the intended agent in **AI Agent Composer**, then select **Agent Details**. The **Model Configuration** section presents four steps: | Field | What to choose | | --------------- | ---------------------------------------------------------------------------------------------------------------------------------------------------- | | **Provider** | Choose the provider for this agent: **Azure OpenAI** and **OpenAI**. | | **Credential** | The intended configured provider connection. A credential name is a label, not a secret to copy into the prompt. | | **Model** | A model available through that provider and credential. Do not assume a model shown in a demo is available in every environment. | | **Temperature** | The response-randomness setting, shown on a scale labelled **Reserved** to **Creative**. Choose it for the task and evaluate the resulting behavior. | When the model configuration shows Completed, test a model response in Playground. Test knowledge retrieval and application skills separately after configuring them. ### Set the agent's identity and instructions The **Agent Configuration** section contains **Agent Name** and **Role Prompt**. Give the agent a name that describes its job and write instructions that cover: * Its purpose and intended users. * Which sources it may use. * When it should ask a clarifying question. * What it should do when information or access is missing. * Whether it may take actions and what confirmation those actions require. #### Illustrative role prompt {% code collapsedlinecount="10" %} ``` You are the Employee Knowledge Assistant. Help employees understand the approved onboarding information connected to this agent. Ask a short clarifying question when the employee's request is ambiguous. Base factual answers on the connected approved sources. If a source is missing or does not answer the question, say so and explain the next step. Do not claim to have opened a ticket or changed an account unless an enabled skill returned evidence of that action. Do not request passwords or other secrets in the conversation. ``` {% endcode %} Adapt this role prompt to your use case. Configure sources, permissions and process approvals separately; the prompt does not create those connections. The screen also includes a prompt selector such as **Default Prompt**, a **+** control, and **Update template**. Check which prompt/template you are editing before updating it; an agent's role instructions and a reusable template should be managed deliberately. ### Save and continue 1. Review the provider, credential, model, and agent instructions. 2. Use **Save** when the edited configuration is ready. 3. Reopen or reread the settings to confirm the intended values are present. 4. Add [knowledge](/userguide/ai-agent-composer/add-knowledge-to-an-agent.md) for answers and [skills](/userguide/ai-agent-composer/configure-agent-skills.md) for application capabilities. 5. [Test in Playground and inspect Logs](/userguide/ai-agent-composer/test-in-playground-and-review-logs.md) before enabling a production channel. If Save is disabled, check whether any field has changed and resolve the validation shown by the form. ### Try it: configure an onboarding-policy agent Build an agent that answers equipment-policy questions from a small synthetic source. This example was tested in presales on 13 October 2026 and uses no application skill or external channel. 1\. Create an agent with a unique test name, such as Docs - Onboarding policy. In Agent Details, select OpenAI and your authorised provider credential. The screenshots use Docs - Onboarding policy - 20261013 and an existing connection named Autom Mate’s Openai; select the connection for your own environment. 2\. Select gpt-4o-mini. The example retained temperature 1.00. These are the sample settings, not a requirement for every agent.
Saved onboarding-policy agent configuration with OpenAI, gpt-4o-mini and temperature 1

Select the provider, credential and model, set the temperature, then save the agent.

3\. Enter this Role Prompt: You are a documentation training assistant. Answer onboarding equipment questions using only the supplied Demo onboarding policy knowledge source. Distinguish employees from contractors. If a detail is absent, say it is not specified. Never claim to order equipment or perform an action. Explain that exceptions go to the demo IT service owner. Keep answers brief. 4\. Use Save for this agent. Do not use Update template for this exercise; it is a separate template action. Reread the saved name, model and prompt before proceeding. 5\. Continue to Add knowledge to an agent and add the small synthetic Text source. Then run the three conversation tests on Test in Playground and review Logs. The configuration was saved and the agent replied in those tests; no application action or external messaging channel was tested.