Configuration

Complete reference for all configuration options in LearnPanta.

Environment Variables

Backend

Frontend

TLDraw Scaffold Worker (Cloudflare)

The scaffold worker uses Wrangler and expects secrets in .dev.vars (local) or Wrangler secrets (prod).

GOOGLE_API_KEY=your_google_api_key
ANTHROPIC_API_KEY=optional
OPENAI_API_KEY=optional

When deploying, run wrangler secret put GOOGLE_API_KEY in tldraw-agent-scaffold/.

Database Configuration

Connection String Format

postgresql://USER:PASSWORD@HOST:PORT/DATABASE

Examples:

# Local development
DATABASE_URL=postgresql://postgres:password@localhost:5432/learnpanta

# Cloud SQL (via proxy)
DATABASE_URL=postgresql://admin:[email protected]:5432/learnpanta

# Cloud SQL (private IP)
DATABASE_URL=postgresql://admin:[email protected]:5432/learnpanta

Connection Pool Settings

SQLAlchemy defaults are used. For production, consider:

# In database.py
engine = create_engine(
    DATABASE_URL,
    pool_size=10,
    max_overflow=20,
    pool_pre_ping=True
)

AI Model Configuration

Gemini API Setup

1. Get API key from Google AI Studio
2. Set GOOGLE_API_KEY environment variable

export GOOGLE_API_KEY=AIzaSy...

Model Selection

Models are configured in backend/app/services/llm.py (override via GEMINI_FLASH_MODEL / GEMINI_PRO_MODEL):

Note: LearnPanta defaults to Gemini 3 across primary workflows. A legacy gemini-2.0-flash fallback exists for compatibility but is not used in the main review pipeline.

Thinking Mode

Thinking mode is available for complex reasoning tasks:

# In llm.py
thinking_config = types.ThinkingConfig(
    thinking_level="high"
)

Currently disabled for performance. Enable for higher-quality reasoning at cost of latency.

Temporal Configuration

Worker Settings

# In worker.py
Worker(
    client,
    task_queue="marathon-session-queue",
    workflows=[MarathonSessionWorkflow],
    activities=[
        incremental_analyzer_agent,
        feedback_synthesis_agent,
        persist_feedback,
        persist_state
    ]
)

Activity Timeouts

Workflow Settings

# History threshold for continue-as-new
if loop_count > 50:
    workflow.continue_as_new(args=[session_id, user_id, self._state])

CORS Configuration

Development

origins = [
    "http://localhost:3000",
    "http://127.0.0.1:3000",
]

Production

Set ALLOWED_ORIGINS environment variable:

ALLOWED_ORIGINS=https://app.yourdomain.com,https://admin.yourdomain.com

Kubernetes Configuration

Secrets

Create Kubernetes secrets for sensitive values:

kubectl create secret generic backend-secrets \
  --from-literal=database-url="postgresql://..." \
  --from-literal=google-api-key="AIzaSy..." \
  --from-literal=api-key="production-secret"

Environment Injection

In deployment YAML:

env:
  - name: DATABASE_URL
    valueFrom:
      secretKeyRef:
        name: backend-secrets
        key: database-url
  - name: GOOGLE_API_KEY
    valueFrom:
      secretKeyRef:
        name: backend-secrets
        key: google-api-key

Resource Limits

Recommended for production:

resources:
  requests:
    memory: "512Mi"
    cpu: "250m"
  limits:
    memory: "2Gi"
    cpu: "1000m"

Environment Profiles

Local Development

• Backend: .env from .env.example; SQLite default (sqlite:///./local_dev.db).
• Frontend: .env.local with NEXT_PUBLIC_API_URL=http://localhost:8000/api/v1, NEXT_PUBLIC_WS_URL=ws://localhost:8000.
• AI: set GOOGLE_API_KEY (backend) and GEMINI_API_KEY (frontend server actions) if testing AI locally.
• Optional: PINECONE_API_KEY for semantic search; TimescaleDB vars if running analytics storage.

Staging

• API/WS URLs point to staging host.
• Use staging Firebase project + TLDRAW key; lower curator/analytics concurrency to control spend.

Production

• Secrets from Secret Manager → Kubernetes secrets; avoid .env files on disk.
• Set ALLOWED_ORIGINS to exact frontend domains (HTTPS + WSS).
• Rotate API_KEY, Firebase credentials, and Gemini keys quarterly; audit Cloud SQL users monthly.

Logging Configuration

Log Level

Set in main.py:

logging.basicConfig(level=logging.INFO)

For debug mode:

logging.basicConfig(level=logging.DEBUG)

Structured Logging

Temporal activities include context:

logger.info(f"Processing session {session_id}", extra={
    "activity_type": "feedback_synthesis",
    "attempt": activity.info().attempt
})

Feature Flags

Currently managed via environment variables:

File Paths

Backend

Frontend

Next Steps

• Deployment - Deploy with these settings
• Troubleshooting - Common configuration issues
• Security - Security best practices