ManageEngine ServiceDesk Plus AI Agent example
Provide a small, useful self-service journey over ServiceDesk Plus: search guidance, create a request, inspect the user's open requests and add a note. Autom Mate Hyperflows perform the API operations; ServiceDesk Plus owns the request records.
Prepare the components
You need AI Agent Composer, a configured AI model, ServiceDesk Plus API access and the relevant credentials in Vault. The documented example uses OpenAI or Azure OpenAI for the model and an API key for ServiceDesk Plus. Select the authentication fields and model/deployment supported by your installed integration.
Scope the integration account to the operations you enable: request creation, request reads, note creation and knowledge reads. If the assistant offers “my requests,” also implement the mapping from the channel user to the ServiceDesk Plus requester. A shared API account alone does not provide per-user filtering.
Configure four focused skills
Create request
Collect required details, map the request payload and return the created record's reference.
“My laptop cannot connect to Wi-Fi.”
View open requests
Identify the user, fetch requests and apply the intended requester/status filters.
“Show my open requests.”
Add note
Resolve the request ID, check access and add the supplied text once.
“Add a note to request 143: it happens only on VPN.”
Search knowledge
Query the configured source and return relevant guidance.
“How do I reset my VPN password?”
In the agent's skills configuration, make each skill's description specific enough to distinguish a read from a write. Define required inputs and preserve the target system's returned identifiers in the result.
Choose a knowledge approach
Use either an API lookup of ServiceDesk Plus knowledge records during execution, or a configured RAG knowledge source that has ingested the required articles. These are alternative designs in the example; configure and test the one you use. Follow knowledge setup and check that returned guidance is available to the requesting user.
Connect and test a channel
Test the agent in Playground. Then configure the chosen channel, such as Teams, Mate Chat or WhatsApp, where supported by your environment. Channel identity and rendering can differ, so repeat the user-specific tests on the final channel.
For the first acceptance test:
Search for a known knowledge article and inspect the returned guidance.
Submit a test request with all required values and inspect it in ServiceDesk Plus.
Ask for open requests as two different test users; confirm each sees the intended records.
Add a note to a permitted test request and verify it appears once.
Try an unknown request number, a denied record and missing input. Check that the response explains the next step without claiming success.
Keep the first release narrow
Status updates, approvals, technician actions and user enrichment are possible extensions to design separately. They are not implied by the four skills above. Once the basic journey is reliable, add one capability at a time with its own input contract, permissions and acceptance tests.
Was this helpful?

