Models & Virtual Aliases

AVAILABLE

Naagmani allows engineering teams to decouple application code from specific underlying provider model strings using Virtual Model Aliases and Smart Routing Policies.

Rendering diagram...

What is a Virtual Model Alias? #

Instead of hardcoding concrete strings like gpt-4o or deepseek-chat across multiple microservices and codebases, your applications request logical aliases defined in your Routing Policies:

json
{
  "model": "smart",
  "messages": [
    {"role": "user", "content": "Analyze quarterly customer retention trends."}
  ]
}

Naagmani intercepts the alias "smart", evaluates the active routing policy for your environment, and dispatches the request to the optimal healthy provider according to the policy rules.


The 3 Model Types Supported in Naagmani #

Naagmani strictly accepts 3 types of model identifiers:

Identifier TypeExampleBehavior
Virtual Alias / Policy Name"smart", "fast", "reasoning", "code-gen"Strictly resolves against your configured Routing Policies. Must be explicitly created in the portal.
Automated Router"auto"Built-in AI routing engine that evaluates all connected providers and models dynamically. (Requires Pro/Enterprise license; blocked on free tier).
Direct Provider Model"deepseek-chat", "gpt-4o", "claude-3-5-sonnet-20241022", "gemini-2.5-flash"Direct dispatch to the specified provider and model registered in your active credential pool.
Note

If a requested model does not match an existing Routing Policy, "auto", or a supported provider model in your pool, Naagmani will immediately reject the request with 400 Bad Request: no eligible provider candidate found for model.


Alias NamePurposeExample Policy Cascade
smart / smart-tierState-of-the-art reasoning and deep analysisDeepSeek R1 / V3 $\rightarrow$ Claude 3.5 Sonnet $\rightarrow$ GPT-4o
fast / fast-tierUltra-low-latency real-time interactionsGemini 2.5 Flash $\rightarrow$ GPT-4o-mini $\rightarrow$ Claude 3.5 Haiku
cheap / economyCost-effective bulk processingDeepSeek V3 $\rightarrow$ Llama 3.3 70B
code / code-genCode synthesis and reviewClaude 3.5 Sonnet $\rightarrow$ DeepSeek Coder

Key Benefits of Virtual Aliases #

  1. Zero-Downtime Model Swaps: When a new flagship model drops (e.g. GPT-5 or DeepSeek V4), switch your smart alias in the Developer Portal without redeploying backend services.
  2. Multi-Provider Resilience: If an upstream provider suffers an outage or rate-limiting (429), your policy automatically cascades to the configured secondary provider.
  3. Environment Isolation:
    • Development/Staging: Point smart to cost-effective models (deepseek-chat or gpt-4o-mini).
    • Production: Point smart to flagship models (claude-3-5-sonnet or deepseek-reasoner).

Next Steps #