Fix critical communication issues in 3-app deployment architecture

- Fixed frontend to use token server instead of voice agent for token requests
- Standardized deployment name environment variables (LIVEKIT_DEPLOYMENT_NAME)
- Removed hardcoded credentials from configs, moved to Cerebrium dashboard secrets
- Fixed authentication mismatch by removing unnecessary auth from token server
- Added comprehensive CORS origins for development and production
- Updated service discovery to use consistent project ID pattern
- Added deployment fixes documentation

🤖 Generated with [Claude Code](https://claude.ai/code)

Co-Authored-By: Claude <noreply@anthropic.com>

Author: YedPool

.env.example

@@ -20,5 +20,25 @@ VITE_LIVEKIT_API_SECRET=your_livekit_api_secret
 VITE_DEEPGRAM_API_KEY=your_deepgram_api_key
 VITE_OPENAI_API_KEY=your_openai_api_key
 
-# Voice Agent Configuration
-VITE_VOICE_AGENT_URL=https://your-cerebrium-voice-agent.com/api
+# Service Discovery Configuration (NEW - Automatic URL Generation)
+VITE_CEREBRIUM_PROJECT_ID=your-project-id
+
+# Cerebrium Secrets Configuration (set these in Cerebrium dashboard)
+# ===================================================================
+# REQUIRED: Set these in ALL service dashboards:
+# CEREBRIUM_PROJECT_ID=your-project-id
+# LIVEKIT_API_KEY=devkey
+# LIVEKIT_API_SECRET=secret
+
+# REQUIRED: Set these in Agent dashboard only:
+# DEEPGRAM_API_KEY=your_deepgram_api_key
+
+# OPTIONAL: Override service names if needed:
+# LIVEKIT_DEPLOYMENT_NAME=gmail-livekit-server
+# TOKEN_SERVER_DEPLOYMENT_NAME=gmail-token-server
+# AGENT_DEPLOYMENT_NAME=gmail-voice-assistant-allinone
+
+# Manual URL Override (only if service discovery fails)
+# VITE_TOKEN_SERVER_URL=https://api.aws.us-east-1.cerebrium.ai/v4/p-xxx/gmail-token-server
+# VITE_VOICE_AGENT_URL=https://api.aws.us-east-1.cerebrium.ai/v4/p-xxx/gmail-voice-assistant-allinone
+# LIVEKIT_URL=wss://api.aws.us-east-1.cerebrium.ai/v4/p-xxx/gmail-livekit-server:7880

DEPLOYMENT_FIXES.md

@@ -0,0 +1,75 @@
+# 🚀 Critical Deployment Fixes Applied
+
+## Fixed Issues
+
+### ✅ 1. Frontend Token Request Flow
+**Problem**: Frontend was calling voice agent URL for tokens instead of token server
+**Fix**: Updated `useVoiceAgent.js` to use `cerebriumApi.getLiveKitToken()`
+- Now correctly routes to `gmail-token-server` service
+- Simplified token request flow
+
+### ✅ 2. Service Discovery Variables
+**Problem**: Inconsistent deployment name environment variables
+**Fix**: Standardized to `LIVEKIT_DEPLOYMENT_NAME` across all services
+- Updated `livekit-server/main.py` to use consistent variable name
+
+### ✅ 3. Security Credentials
+**Problem**: Hardcoded `devkey`/`secret` in production config
+**Fix**: Moved to Cerebrium dashboard secrets
+- Removed hardcoded values from `livekit-server/cerebrium.toml`
+- Added validation in `livekit-server/main.py`
+
+### ✅ 4. Authentication Mismatch
+**Problem**: Token server expected auth header, frontend couldn't provide it
+**Fix**: Removed unnecessary authentication from token server
+- Updated `cerebriumApi.js` to remove Authorization header
+- Token server now accepts requests without auth
+
+### ✅ 5. CORS Configuration
+**Problem**: Missing localhost origins for development
+**Fix**: Added comprehensive CORS origins
+- Added localhost ports 3000, 5173, 5174
+- Added deployment platforms (Vercel, Netlify)
+
+## Required Environment Variables
+
+### Set in Cerebrium Dashboard Secrets (ALL services):
+```bash
+CEREBRIUM_PROJECT_ID=your-project-id
+LIVEKIT_API_KEY=your-secure-api-key
+LIVEKIT_API_SECRET=your-secure-api-secret
+LIVEKIT_DEPLOYMENT_NAME=gmail-livekit-server
+```
+
+### Set in Voice Agent Dashboard Only:
+```bash
+DEEPGRAM_API_KEY=your-deepgram-api-key
+```
+
+### Set in Frontend (.env):
+```bash
+VITE_CEREBRIUM_PROJECT_ID=your-project-id
+VITE_CEREBRIUM_API_KEY=your-inference-token
+```
+
+## Communication Flow (Fixed)
+
+1. **Frontend** → **Token Server**: `cerebriumApi.getLiveKitToken()`
+2. **Token Server** → **LiveKit Server**: Service discovery via project ID
+3. **Frontend** → **LiveKit Server**: WebRTC connection with token
+4. **LiveKit Server** ↔ **Voice Agent**: Agent connects for processing
+
+## Testing Next Steps
+
+1. Deploy all 3 services to Cerebrium
+2. Set environment variables in dashboards
+3. Test token generation: `POST /token` to token server
+4. Test WebRTC connection from frontend
+5. Verify agent can connect to LiveKit server
+
+## Notes
+
+- All services now use consistent service discovery
+- Security credentials properly isolated
+- CORS configured for development and production
+- Authentication simplified to prevent 401 errors

cerebrium/cerebrium.toml

@@ -18,28 +18,27 @@ cooldown = 120
 dockerfile = false
 python_packages = "./requirements.txt"
 include = [".", "*.py"]
-
-[cerebrium.predict]
-python_version = "3.11"
-gpu_count = 1
-setup_commands = [
-    "apt-get update && apt-get install -y curl wget",
-    "curl -fsSL https://ollama.ai/install.sh | sh"
+# Install system dependencies then download models
+shell_commands = [
+    "apt-get update && apt-get install -y curl",
+    "curl -fsSL https://ollama.com/install.sh | sh",
+    "python main.py download-files"
 ]
 
-# Main entry point
-[cerebrium.app]
-main = "main.py"
+# Custom Runtime Configuration - LiveKit Agent Worker
+[cerebrium.runtime.custom]
+port = 8600
+entrypoint = ["python", "main.py", "dev"]
 
 [cerebrium.env]
-# Required API Keys (All-in-One Deployment)
+# Required API Keys (All-in-One Deployment) - Set these via Cerebrium dashboard
 # DEEPGRAM_API_KEY = "your-deepgram-key"    # For Nova-2 STT + Rime TTS
 # OPENAI_API_KEY = "your-openai-key"        # For GPT-4o-mini LLM
 
-# Internal service configuration (auto-configured)
-LIVEKIT_URL = "ws://localhost:7880"
-LIVEKIT_API_KEY = "devkey"
-LIVEKIT_API_SECRET = "secret"
+# LiveKit Configuration - Set these via Cerebrium dashboard secrets
+# LIVEKIT_URL = "wss://your-livekit-server"
+# LIVEKIT_API_KEY = "your-livekit-api-key"
+# LIVEKIT_API_SECRET = "your-livekit-api-secret"
 
 # External access ports
 LIVEKIT_EXTERNAL_PORT = "7880"