Agent Setup

Configure Rian, your AI agent, from onboarding through to going live.

Renprofile's AI agent is called Rian. Setup happens in two phases: a short strategy questionnaire during onboarding, followed by configuration and testing inside the workspace.


Phase 1: Onboarding strategy

After you complete workspace setup and team invites, Renprofile walks you through a 4-step agent strategy flow. Your answers shape how Rian behaves from day one.

Step 1: Who are your customers?

Select the audience profile that best describes your users:

OptionDescription
B2B SaaS usersBusiness teams using your software product
DevelopersTechnical users building with your platform
Non-technical business usersOps, finance, or other non-dev business roles
ConsumersEnd consumers using a product or service

Step 2: What language do your customers speak?

Option
English (Default)
Spanish
French
German
They Speak Multiple Languages

Step 3: What should Rian never handle?

Select one or more topics Rian should always escalate to a human agent. You can select multiple:

Topic
Billing disputes
Legal questions
Enterprise account issues
Refund requests
(None — Rian handles everything it can)

Step 4: What does a good resolution look like?

This tells Rian what success means for your team:

OptionMeaning
Customer resolved without a humanRian should aim to fully self-serve the customer
Customer was routed to the right person quicklyRian should prioritise fast, accurate handoff

Once you complete step 4, Renprofile begins ingesting your knowledge base and takes you to the workspace.


Phase 2: Configuration

After onboarding, further agent configuration is available from Settings > AI Agent. This is where you can update the strategy answers, adjust escalation rules, and manage the knowledge base.


Phase 3: Sandbox

Before going live, test Rian in the Agent Sandbox at Workspace > Sandbox. The sandbox has two panels:

Left panel — debug view: three tabs give you a live look at what Rian is doing under the hood.

TabWhat it shows
SessionThe message input, plus Classification and Flow State data after each simulation
EventsA timeline of internal events: classification, retrieval, LLM call, and transition — each with a latency reading
MemoryFour memory tiers (Hot, Working, Warm, Cold) with fetch latency and expandable key/value snapshots

Classification data (Session tab):

FieldDescription
Intent ClassThe broad category of the message
IntentThe specific intent detected
ScenarioThe matched scenario
ConfidenceScore from 0.0 to 1.0; above 0.8 is highlighted green
Intent lockedTrue when confidence exceeds 0.9

Flow State data (Session tab):

FieldDescription
Active stateThe current flow being executed
Active stepThe step within that flow
AttemptsHow many times the step has been tried
OutcomeResult: resolved, escalated, or pending

Right panel — chat preview: a live Beacon widget simulation. Select who you're simulating as before sending a message:

Simulate asWhat it represents
New VisitorAnonymous user with no history
Returning UserUser seen before but not identified
Known CustomerIdentified user with stored attributes

Use Reset Conversation to clear the session and start fresh.


Go live

Once you're satisfied with sandbox results, enable Rian on live traffic from Settings > AI Agent > Go Live. Rian will immediately begin handling new conversations on all connected channels.

On this page