Mode Selection FAQ

Mode Selection FAQ

General Questions

Q: What’s the main difference between On-Demand and Server modes?

A: On-Demand mode runs as a command-line tool that executes and exits, while Server mode runs continuously as a service. Think of On-Demand as a calculator you pick up when needed, and Server mode as a dedicated assistant always ready to help.

Q: Can I switch between modes easily?

A: Yes! The same installation supports both modes. You can use On-Demand for development and Server for production, or switch based on your current needs. Data is compatible between modes.

Q: Which mode is better for beginners?

A: On-Demand mode is generally better for beginners because:

  • Simpler mental model (run command → get result)
  • No process management required
  • Easier to debug and understand
  • Lower complexity overall

Q: Do both modes use the same AI providers?

A: Yes, both modes support the same providers (OpenAI, Anthropic, Google, Ollama, OpenRouter) and use the same configuration format.

Performance Questions

Q: How much faster is Server mode really?

A: For individual requests:

  • On-Demand: 100-500ms startup + processing time
  • Server: 0ms startup + processing time

For 100 requests:

  • On-Demand: ~30-50 seconds (sequential)
  • Server: ~5-10 seconds (with caching and parallelism)

Q: When does the performance difference matter?

A: Server mode performance benefits are significant when:

  • Making more than 10 requests per minute
  • Building interactive applications
  • Need sub-second response times
  • Running batch operations frequently

Q: What about memory usage?

A:

  • On-Demand: 0MB when idle, 50-100MB when running
  • Server: 50-200MB constant (increases with cache size)

For reference, 200MB is less than a single Chrome tab.

Feature Questions

Q: What features are exclusive to Server mode?

A: Server mode exclusive features:

  • Real-time adaptive learning
  • WebSocket support
  • Session management
  • Background optimization
  • In-memory caching
  • Pattern recognition
  • Multi-user support
  • Hot configuration reload

Q: Can On-Demand mode learn from usage?

A: Limited learning is possible through:

  • Usage statistics saved to database
  • Manual feedback recording
  • Batch analysis of results

However, real-time adaptive learning requires Server mode.

Q: Is the output quality different between modes?

A: No, the core prompt generation engine is identical. Server mode may provide better results over time through learning, but initial quality is the same.

Integration Questions

Q: Which mode works better with CI/CD?

A: On-Demand mode is ideal for CI/CD because:

  • Clean process lifecycle
  • Exit codes for success/failure
  • No persistent state required
  • Easy to containerize
  • Simple resource management

Q: Can I use Server mode in serverless functions?

A: Not recommended. Serverless is better suited for On-Demand mode. For serverless with Server mode benefits, consider:

  • API Gateway → Server mode instance
  • Edge functions → Server mode endpoint
  • Managed container services

Q: How do I integrate with VS Code or other IDEs?

A: Both modes work, but differently:

  • On-Demand: IDE spawns process for each operation
  • Server: IDE maintains connection to running server

Server mode is generally better for IDE integration due to lower latency.

Deployment Questions

Q: Which mode is easier to deploy?

A: On-Demand mode is simpler:

  • Copy binary → done
  • No service management
  • No port configuration
  • No monitoring setup

Server mode requires additional setup but provides better operational visibility.

Q: Can I run multiple Server instances?

A: Yes, with considerations:

  • Use shared storage for learning data
  • Configure session affinity for WebSockets
  • Implement cache synchronization
  • Use external load balancer

Q: What about security?

A: Security comparison:

Aspect On-Demand Server Mode
Network exposure None Configurable
Authentication OS-level Application-level
Attack surface Minimal Requires hardening
Secrets management Environment Multiple options

Cost Questions

Q: Which mode is more cost-effective?

A: Depends on usage patterns:

On-Demand is cheaper when:

  • < 100 requests/day
  • Sporadic usage
  • Running on personal machines
  • Shared infrastructure

Server mode is cheaper when:

  • 1000 requests/day

  • Continuous usage
  • Dedicated infrastructure
  • Need caching benefits

Q: How do API costs compare?

A: API costs are similar, but Server mode can reduce them through:

  • Intelligent caching
  • Request deduplication
  • Batch processing
  • Learning-based optimization

Migration Questions

Q: How do I migrate from On-Demand to Server?

A: Simple process:

  1. Export data: prompt-alchemy export
  2. Update config to enable server mode
  3. Start server: prompt-alchemy serve
  4. Update integrations to use MCP client
  5. Import data if needed

Q: What about migrating from Server to On-Demand?

A: Also straightforward:

  1. Export learned patterns via API
  2. Stop server gracefully
  3. Update scripts to use CLI
  4. Continue with same database

Q: Can I run both modes simultaneously?

A: Yes, but:

  • Use different ports for Server mode
  • Be aware of potential database conflicts
  • Consider read-only mode for one instance
  • Synchronize configuration changes

Troubleshooting Questions

Q: My On-Demand commands are slow, should I switch to Server?

A: First try:

  • Check API key validity
  • Verify network connectivity
  • Use --no-update-check flag
  • Ensure database isn’t corrupted

If consistently slow with many requests, then consider Server mode.

Q: Server mode memory usage keeps growing, is this normal?

A: Some growth is normal due to:

  • Learning pattern accumulation
  • Cache expansion
  • Session storage

Configure limits:

cache:
  max_size: 1000
  ttl: 3600
learning:
  pattern_limit: 10000
  cleanup_interval: 6h

Q: Which mode is better for debugging?

A: On-Demand is generally easier to debug:

  • Clear start/stop boundaries
  • Simple stdout/stderr
  • No concurrent operations
  • Easier to reproduce issues

Server mode provides better observability through metrics and structured logs.

Best Practices Questions

Q: Should I use Server mode in development?

A: Consider your workflow:

  • Frequent prompt generation → Server mode
  • Occasional usage → On-Demand
  • Team development → Server mode (shared instance)
  • Personal projects → Either works

Q: How do I choose for production?

A: Production decision factors:

Choose On-Demand if:

  • Batch processing workflows
  • Scheduled jobs (cron)
  • Simple integrations
  • Resource constraints

Choose Server if:

  • User-facing applications
  • Real-time requirements
  • Need learning capabilities
  • Multiple integrations

Q: Can I start with one mode and switch later?

A: Absolutely! This is the recommended approach:

  1. Start with On-Demand for simplicity
  2. Evaluate usage patterns
  3. Switch to Server when benefits justify complexity
  4. Keep On-Demand for scripts/automation

Advanced Questions

Q: How does learning work differently between modes?

A: Learning comparison:

Feature On-Demand Server Mode
Usage tracking ✅ Basic ✅ Comprehensive
Pattern detection ❌ ✅ Real-time
Adaptation ❌ ✅ Continuous
Feedback loop Manual Automatic
Optimization ❌ ✅ Background

Q: What about horizontal scaling?

A: Scaling strategies:

On-Demand:

  • Scale by running parallel processes
  • Use job queues (Celery, Sidekiq)
  • Distribute via container orchestration

Server Mode:

  • Built-in concurrent request handling
  • Support for load balancer health checks
  • Session affinity for learning context
  • Shared cache layer (Redis)

Q: Which mode is better for compliance/auditing?

A: Both support auditing, differently:

On-Demand:

  • Process-level audit trails
  • Simple log correlation
  • Clear execution boundaries

Server Mode:

  • Centralized logging
  • Request tracing
  • Comprehensive metrics
  • Session tracking

Choose based on your compliance requirements.

Still have questions? Check the comprehensive comparison or quick reference.