Sandbox

The Agent Sandbox lets you simulate conversations with Rian in a safe environment. No live traffic is affected.

Go to AI Agent > Sandbox in the sidebar.

Layout

The sandbox has two panels side by side.

Left panel: debug view

Three tabs give you a live look at what Rian is doing under the hood:

Tab	What it shows
Session	Message input; Classification data; Flow State data
Events	Timeline of internal events with latency readings
Memory	Four memory tiers with fetch latency and key/value snapshots

Session tab: Classification

Field	Description
Intent Class	Broad category of the message
Intent	Specific intent detected
Scenario	Matched scenario
Confidence	Score 0.0–1.0; above 0.8 highlighted green
Intent locked	True when confidence exceeds 0.9

Session tab: Flow State

Field	Description
Active state	Current flow being executed
Active step	Step within that flow
Attempts	How many times the step has been tried
Outcome	resolved; escalated; or pending

Events tab

Four events are tracked per simulation: classification; retrieval; LLM call; transition. Each shows a status dot (green for success, blue for active) and a latency reading in milliseconds.

Memory tab

Four tiers are shown, each expandable to reveal key/value data:

Tier	Badge	Description
Hot	in use	Active session context
Working	active	Short-term working memory
Warm	cached	Recently accessed knowledge
Cold	fetched	Knowledge base retrieval

Right panel: chat preview

A live Beacon widget simulation. Select who you're simulating as before sending a message:

Option	Represents
New Visitor	Anonymous user with no history
Returning User	User seen before but not identified
Known Customer	Identified user with stored attributes

Click Reset Conversation to clear the session and start fresh. The chat supports the same message input as the live widget including Enter to send.