Models & Virtual Aliases

Naagmani allows engineering teams to decouple application code from specific underlying model strings using Virtual Model Aliases.

What is a Virtual Model Alias? #

Instead of hardcoding gpt-4o or claude-3-5-sonnet-20241022 into your backend repositories, your applications request virtual tier aliases:

graph LR
    App["Application Code"] -->|model: smart-tier| Router["Naagmani Router"]
    Router -->|Rule: Primary| C35["Claude 3.5 Sonnet"]
    Router -->|Rule: Fallback| G4O["OpenAI GPT-4o"]

Common Alias Patterns: #

• smart-tier: Resolves to state-of-the-art reasoning models (e.g., Claude 3.5 Sonnet $\rightarrow$ GPT-4o).
• fast-tier: Resolves to ultra-low-latency, lightweight models (e.g., Gemini 2.5 Flash $\rightarrow$ GPT-4o-mini).
• cheap-tier: Resolves to high-throughput, cost-efficient models (e.g., DeepSeek V3).

Benefits of Virtual Aliases #

1. Instant Model Upgrades: When a new model version is released (e.g. GPT-5), update the alias target in the Developer Portal with zero application downtime or deployments.
2. Zero Code Refactoring: Switch primary providers across hundreds of microservices instantly.
3. Environment-Specific Routing: Route smart-tier to an inexpensive model in Test and the premier flagship model in Production.

Next Steps #

• Configure routing rules: Smart Model Routing
• Test models interactively: Model Playground