feat: Portal Chat - Implement controlled demo pricing model

Summary

Portal chat currently uses admin API key (CONTROL_LAYER_ADMIN_API_KEY) which means users are NOT charged for chat usage. This needs to be fixed to use proper billing while positioning chat as a demo/onboarding tool.

Current Problem

// chat.controller.ts:364-365
const adminApiKey = process.env.CONTROL_LAYER_ADMIN_API_KEY || '';
const completion = await controlLayer.createCompletion({...}, adminApiKey);

• All chat costs go to admin/system account
• User's balance is NOT affected
• ChatConversation.totalCost is just metadata, not actual billing

Proposed Solution

Positioning

Chat is not a product - it's a demo instrument for:

• Lowering threshold to understand Qwen3 30B VL (text + vision)
• Showing latency, quality, multimodality without SDK
• Reducing support friction during onboarding

Pricing Model

• 10% of standard token price (not a discount - it's positioning)
• Same wallet/credits as API
• All usage visible in usage dashboard

Technical Implementation

1. Use user's API key (not admin key)

• Require user to select/create an API key for chat
• Or auto-create a "portal-chat" key per org

1. Apply chat discount in Control Layer

• New pricing tier: chat_demo at 10% of standard
• Or apply multiplier in Portal BFF before calling CL

1. Guardrails

• Max context: 16k tokens
• Rate limit per user/session
• No streaming burst
• No batch operations

UI Changes

• Label chat as: "Interactive demo — not intended for production workloads"
• Position as "Try the model" not "Chat with AI"
• Signal: "Inference provider, not chat app"

Files to Modify

• portal-bff/src/controllers/chat.controller.ts - Use user API key, apply pricing
• portal-bff/src/controllers/streaming-chat.controller.ts - Same changes
• portal-ui/app/(dashboard)/chat/page.tsx - Add demo disclaimer
• Control Layer: Add chat_demo pricing tier (optional)

Acceptance Criteria

• [ ] Chat uses user's org balance, not admin account
• [ ] Chat priced at 10% of standard API rate
• [ ] Clear "demo/playground" labeling in UI
• [ ] Rate limits and context limits enforced
• [ ] Usage appears in user's usage dashboard

Priority

Medium - Current setup works but has incorrect billing model

---
Created from portal code analysis 2026-01-11