copilotkit--copilotkit/showcase/integrations/langroid/qa/agent-config.md

QA: Agent Config Object — Langroid

Prerequisites

• Demo is deployed and accessible
• Agent backend reachable at /api/copilotkit-agent-config
• Langroid agent server running (see /api/health)

Test Steps

1. Config UI

• Navigate to /demos/agent-config
• Verify data-testid="agent-config-card" is visible
• Verify Tone / Expertise / Response length selects are rendered

2. Forwarded properties reach the agent

• Change Tone to "enthusiastic"
• Send "Hello" and verify a response is produced; tone should read as noticeably upbeat/warm
• Change Expertise to "expert" and Response length to "detailed"
• Send "Explain how LLM tool calling works" — verify the response uses domain terminology freely and is multiple sentences (not 1 to 2)
• Change Response length to "concise" and Expertise to "beginner"
• Send the same question — response should be 1 to 2 sentences, avoid jargon, and define any technical term the first time it appears

3. Network inspection (optional, deeper verification)

• Open DevTools Network panel
• Send a message and inspect the POST to /api/copilotkit-agent-config
• In the request body, verify forwardedProps.config.configurable.properties contains tone, expertise, and responseLength with the selected values
• The flat keys forwardedProps.tone / .expertise / .responseLength should NOT be present — the route repacks them under config.configurable.properties

Expected Results

• Selecting different config values visibly changes the assistant's voice, depth of explanation, and response length.
• The /api/copilotkit-agent-config request body shows the repacked shape (flat provider keys land under forwardedProps.config.configurable.properties).
• The Langroid backend receives the properties (via AG-UI forwarded_props) and appends style directives to its system prompt for that run only; other demos remain unaffected.