Sign up, choose an agent template, add one policy source, and verify three answers before expanding to customers.
WORK INResponseCX app
BRING
Workspace access · One policy · A reviewer
FINISH WITH
A saved agent and three reviewed answers
Outcome: one agent answers a known question, asks for missing information, and escalates an
unsupported request. You will use the ResponseCX app; no API key is required for this walkthrough.
Agent creation and imported sources belong to that workspace
One current shipping, returns, or support policy
Gives your reviewer an authoritative answer to compare against
Three test questions and a named reviewer
Checks a known answer, missing context, and a policy boundary
Finish line: a saved agent, a processed source, and three reviewed answers. Publishing a
customer-facing channel is a separate next step. Keep the agent ID and results in your
onboarding record.
Sign up or sign in to your existing organization, then
open Setup. Follow any organization or plan steps shown.
If you came from the provisioning portal, use Open your Response workspace after launch.Check the selected organization before importing knowledge or creating an agent. Organization
identity in Response and a workflow brand UUID are different identifiers; retain both when you
later connect workflow automation.
In Choose agent, select the template closest to your first task and click Create my agent.
The setup templates seed instructions, rules, attributes, examples, and model defaults. The
confirmation shows what was created; Edit in Agents opens the management area.Choose one narrow first task, such as explaining your published shipping policy. Do not add
refund execution, order changes, and every channel to that first test.
Success: the setup screen confirms the agent exists, and you can find it in
Agents.
The wizard’s essential steps are Choose agent and Test & launch. Add sources,
Connect systems, Automate your queue, and Invite team are additional setup steps.
For a grounded first answer, use Add sources first on the no-knowledge notice before testing.Import a relevant website or document through Add sources. Wait for processing to finish,
then inspect it in Knowledge Base. Sources belong
to the organization; importing one is not a private upload to only the agent on this screen.
Success: the source is processed and its policy text matches the policy you intend to apply.
If you have no source yet, the agent can use its instructions, but do not describe that as a
knowledge-grounded answer.
Return to Test & launch and use the chat widget on the page. This is the live widget runtime;
generated answers use model resources and count toward usage. If setup is blocked by evaluations,
use Review evaluations to inspect the failed checks.
Verify: The agent follows configured policy and escalation behavior instead of promising an unauthorized exception.
Save the observed answers and what you checked. These questions are test inputs, not a promise
that an unconfigured template will pass. Correct the source, instructions, or rules and rerun a
failed case. Use Evaluations to preserve regression cases.
First value: you have an answer your team can verify, plus evidence that the agent handles
missing information and policy boundaries appropriately.
The setup screen provides the widget installation snippet and links to
Widget Builder. Install the generated snippet
on the intended site and test it there. Use the app’s Connect systems step for the integrations
your selected channel or use case needs.Review the agent’s permitted actions and autonomy settings before exposing it to customer
traffic. Creating an agent, publishing a widget, connecting a helpdesk, and switching workflow
routing live are different actions; completion of one does not establish the others.For a complete website handoff test, follow Widget to desk.
Use Chat Desk to coordinate operators and inspect supported replies;
check channel coverage before adding another channel.
Before leaving, update your onboarding record
with the identifiers, observed result, remaining blocker, and next action. If you switch to an
AI assistant, use the resume prompt
to verify current state before continuing.