Model Migration

Overview

Allternit’s LLM gateway provides a unified API across multiple model providers. Switching models — or running multiple models in parallel — requires no code changes, just configuration.

Supported providers

Provider	Models	How to connect
Allternit Gateway	Policy aliases and connected models	Default
Local (Ollama)	Any GGUF/MLX model	Connect via Providers API
Commercial API providers	Provider-native model catalogs	Connect via Providers API

Using policy aliases

Policy aliases are the simplest way to migrate. The gateway resolves an alias to a concrete provider/model pair and derives a cross-provider fallback chain automatically.

Alias	Optimization target
`auto`	Same as `allternit-balanced`.
`allternit-balanced`	Balanced quality and cost across all benchmarks.
`allternit-code`	Coding and software-engineering tasks.
`allternit-reasoning`	Reasoning, math, and science tasks.
`allternit-knowledge`	Knowledge and recall tasks.
`allternit-instruct`	Instruction-following and chat quality.

const response = await client.chat.completions.create({
  model: 'allternit-code',
  messages: [{ role: 'user', content: 'Refactor this function' }],
});

Using explicit provider/model ids

For deterministic routing, pass a connected provider and model directly:

const response = await client.chat.completions.create({
  model: 'provider-id/model-id',
  messages: [{ role: 'user', content: 'Hello' }],
});

Migration checklist

List connected models with GET /v1/models.
Test with a small workload using a policy alias or explicit model.
Compare outputs, latency, and cost via the Estimates API and Usage logs.
Update your application to use the new model string.
Monitor usage via the audit log for the first 24 hours.

Cost comparison

Use the cost estimation API before migrating:

curl -X POST https://api.allternit.com/v1/estimates/cost \
  -H "Authorization: Bearer $ALLTERNIT_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "allternit-code",
    "messages": [
      {"role": "user", "content": "Refactor this function"}
    ],
    "max_tokens": 500
  }'

Prompt caching

Allternit supports prompt caching for models that advertise it. Cached prompts reduce latency and cost for repeated system prompts:

const response = await client.chat.completions.create({
  model: 'allternit-balanced',
  messages: [
    {
      role: 'system',
      content: 'You are a code reviewer...',
      cache_control: { type: 'ephemeral' }
    },
    { role: 'user', content: diff }
  ],
});

For caching entire message arrays, see Context Caches.

Reasoning mode

Enable extended thinking for complex tasks:

const response = await client.chat.completions.create({
  model: 'allternit-reasoning',
  messages: [{ role: 'user', content: 'Prove this theorem' }],
  reasoning_effort: 'high'
});

Get Started

Core Concepts

Gizzi Code CLI

Tools & ACI

Security & Enterprise

Admin API

BYOC Runtime

Surfaces

CI CD

Guides

Cookbooks

Release Notes

Overview

Supported providers

Using policy aliases

Using explicit provider/model ids

Migration checklist

Cost comparison

Prompt caching

Reasoning mode

​Overview

​Supported providers

​Using policy aliases

​Using explicit provider/model ids

​Migration checklist

​Cost comparison

​Prompt caching

​Reasoning mode

Overview

Supported providers

Using policy aliases

Using explicit provider/model ids

Migration checklist

Cost comparison

Prompt caching

Reasoning mode