How ServiceNow AI-agent RAG works
ServiceNow's Search retrieval tool adds RAG to an AI agent in AI Agent Studio. An administrator selects a search profile and sources, chooses fields to return and a result limit, and selects semantic or keyword criteria. The tool description is also sent to the model, so its scope needs plain, specific wording. ServiceNow documents sn_aia.admin as the required role to add it.
Behind that tool is AI Search. ServiceNow describes AI Search RAG as indexing data and then searching that indexed material. It uses an embedding model for semantic or vector search. The practical consequence is simple: an agent can only retrieve what was indexed, permitted, and described well enough for the query and tool to use.
The retrieval-tool form includes a description, search profile, result fields, and search criteria. Source: ServiceNow Search retrieval.
Test the evidence, not just the answer
Take a support policy that changed after a product release. Build a small test set: a normal request, a question whose older policy must not win, missing evidence, a confidential record, and an exception needing an owner. For each case, inspect the retrieved material, the final wording, and any action request separately.
If the agent retrieves old material, correct the source or profile and test again. If it retrieves the right source but offers an unauthorized exception, the retrieval layer is not the issue: the instruction, action boundary, or approval process is. Before production, include a failure path and a handoff. After a controlled release, check the actual case or downstream record, not merely an articulate response.
Inspect an eesel knowledge setup from a terminal
For an eesel helpdesk teammate, eesel CLI gives a support owner, script, or coding agent a terminal path into the same workspace and dashboard used by support. It is useful for a source-review note, while ServiceNow's Search retrieval tool remains the way to configure RAG inside ServiceNow. Claude Code, Codex, and Cursor can read the CLI's JSON output for a scoped review.
For example, when a returns-policy document changes, a coding agent can inspect the named eesel teammate’s connected sources, manually uploaded files, and instructions. The work is to identify whether the approved document and the handoff rule are present, not to connect a new account or turn on an action during inspection.
With Node.js 18.17 or newer, make the target explicit:
npx @eesel/cli login
npx @eesel/cli whoami
TARGET_EESEL_AGENT="paste-agent-id-or-name-here"
npx @eesel/cli integrations --agent "$TARGET_EESEL_AGENT"
npx @eesel/cli files ls --agent "$TARGET_EESEL_AGENT"
npx @eesel/cli instructions --agent "$TARGET_EESEL_AGENT"
After the owner approves an updated file, run --dry-run on the supported upload before sending it. Contain actions outside the test before asking a representative question. Create a named chat, check the cited policy and handoff, then inspect activity and the actual ticket or connected record after controlled rollout. Chat is billed work; dashboard-management commands are free. JSON makes the review repeatable, but it does not replace approval or a result check.
Use retrieval as a bounded capability
The strongest RAG setup is not the one with the most sources. It has a defined job, a source owner, access boundaries, and questions that prove what happens when evidence is insufficient. Apply the same discipline to ServiceNow Search retrieval and to an eesel support teammate.
Review knowledge before expanding automation
The eesel Activity page shows a work list and filters for approved, rejected, and pending items.
Try eesel to review a specific support teammate’s knowledge before it answers more customer cases.
Frequently asked questions
What is RAG for an AI agent?
Retrieval-augmented generation retrieves relevant information from a configured dataset and supplies it as context for a model response. It helps ground an answer, but teams still need to manage source freshness, access, and what the agent may promise or do.
How does ServiceNow add RAG to an AI agent?
ServiceNow's Search retrieval tool can be added in AI Agent Studio. An administrator chooses a search profile, search sources, returned fields, result limit, and semantic or keyword criteria.
Who can configure Search retrieval?
ServiceNow documents the sn_aia.admin role as required to add Search retrieval. Available features and content depend on the instance’s configuration and access controls.
What should teams test?
Test a correct source, a stale source, missing or conflicting evidence, unauthorized data, and a question that needs a human decision. Inspect the retrieved evidence and the final response separately.
Does retrieval guarantee a correct answer?
No. Retrieval can return irrelevant, incomplete, or outdated material. A grounded response still needs response rules, human handoff, and checks of the system where work is completed.
