Skip to main content

Agent Test Panel

The test panel lets you chat with an agent directly from the admin panel and inspect every execution detail without deploying to a channel first.

Accessing the Test Panel​

From the agent list, click the flask icon on any agent card. Or navigate to /agents/:id/test directly. The agent edit page also has a Test button in its sticky footer.

The layout shows two tabs at the top of every agent page: Edit and Test. Switch between them without losing your place.

Chat Interface​

The test panel is a full chat interface connected to your live agent configuration. Messages are sent to POST /tenants/:slug/agents/:agentSlug/run with a persistent session_id, so the conversation maintains context across messages within the same test session.

Each message bubble shows its routing path: Rule (deterministic fast path) or LLM (model call). Click any message to open the inspector.

Inspector Modal​

The inspector modal has four tabs:

Summary​

High-level metrics for the selected run:

  • Token counts (input / output)
  • Step count
  • Latency in milliseconds
  • Run ID and Session ID with one-click copy buttons

Flow​

An n8n-style diagram showing the execution path:

User → Agent → [Tool A] → [Tool B] → Response
               [Tool C]

Tool names are humanized (e.g. fyso_data becomes "Fyso Data"). The diagram reflects the actual tools called during that run, not all available tools.

Steps​

Full message history split into two sections:

  • History — turns from previous messages in the session
  • Current turn — the messages exchanged for this specific run, including tool call requests, tool results, and the final assistant response

Raw​

Complete JSON payload for the run. Useful for debugging unexpected behavior or verifying exact token counts and metadata.

Agent List Quick Actions​

Each agent card in the list now shows two quick-action buttons:

ButtonAction
Flask iconOpen test panel for this agent
List iconOpen rules editor for this agent

Data Access Scope​

The agent form (Edit tab) now includes a Tools scope section where you select which entities the agent can access when using data tools. Changes here take effect immediately on the next test run.