Model and instructions
Select an available model through the configured OpenAI or Azure OpenAI connection. Set the role prompt and response behavior.
FROM THE WORKSHOP
Explore automation, service delivery and the next thing to build.
Browse all articles ↗AI AGENT COMPOSER
Configure what the agent knows, which actions it can use and where people interact with it. Test its behavior before opening the conversation to users.
CONFIGURE
Start with the agent's purpose, then connect the sources and actions that support it.
Select an available model through the configured OpenAI or Azure OpenAI connection. Set the role prompt and response behavior.
Add files, public websites, text or an available application source. Check answers against the information you connected.
Choose an application action and its authorized connection. Configure the inputs and check the returned result.
Connect Microsoft Teams, WhatsApp, Mate Chat or a webhook, using the setup required for that destination.
PLAYGROUND AND LOGS
Use Playground for controlled conversations. Review messages and replies in Logs, then inspect the underlying Hyperflow when an application action needs investigation.
Check known answers, ambiguous requests and information the agent does not have.
Compare a skill's response with the destination record when the action changes data.
Confirm that the intended user reaches the agent and receives its reply in the actual destination.
CUSTOMER SCOPE
In an MSP environment, AI agents and Vault records belong to a tenant. Select the intended tenant and authorized application connection when configuring a skill.
A reference document supplies information. An application skill supplies a capability. Configure and test each separately.
Composer skills and tools exposed to an external AI client through MCP are separate interfaces. Use the relevant setup guide for each.
Model, instructions and the six Composer tabs.
Read the configuration guideConnect application capabilities and choose the interaction channel.
Open Composer documentationTest behavior and trace conversation evidence.
Read the testing guideReview its knowledge, application access and human handover with our team.