Routing LiteLLM to Claude

LiteLLM is the proxy layer many teams put in front of everything, exposing one OpenAI-compatible API to all their apps. Point its model_list at AI Prime Tech and that single endpoint reaches Claude — plus GPT and Gemini through the same multi-model gateway.

model_list and the provider prefix

In config.yaml each model_list entry has a client-facing model_name and litellm_params. The model field carries a provider prefix that decides the wire protocol: anthropic/<model> sends the native Messages API, openai/<model> sends Chat Completions.

Set api_base to the gateway and api_key via os.environ/<VAR>. Because AI Prime Tech speaks the OpenAI-compatible surface, openai/<claude-model> with an api_base ending in /v1 is the simplest route.

The URL-suffix gotcha

With anthropic/<model>, LiteLLM auto-appends /v1/messages to the root api_base. To control an exotic path, set LITELLM_ANTHROPIC_DISABLE_URL_SUFFIX=true and supply the full path. With openai/<model>, give the base with /v1 and LiteLLM appends /chat/completions.

The prefix also dictates the JSON body shape, so a mismatch with what the gateway expects returns 400s — keep prefix and gateway surface aligned.

LiteLLM as the unifying layer

The classic pattern is LiteLLM as a local proxy that Claude Code, scripts, and other tools all target. You swap upstream models or gateways in config.yaml without touching every client, and get unified logging, routing, and fallbacks.

With AI Prime Tech upstream, one LiteLLM endpoint fans out to Claude, GPT, and Gemini, so your whole stack speaks one API while you keep full model choice.

Where volume concentrates

LiteLLM is where many apps converge, so it carries the densest request stream in the stack. That makes it the place where cost predictability matters most.

A multi-model gateway behind it means you can route cheap models for bulk jobs and Claude for the hard ones, all measured in one place.

# config.yaml
model_list:
  - model_name: claude-sonnet-4-5
    litellm_params:
      model: openai/claude-sonnet-4-5
      api_base: https://aiprimetech.io/v1
      api_key: os.environ/AIPRIME_KEY
# litellm --config config.yaml

Frequently asked questions

How do I point LiteLLM at AI Prime Tech?
Add a model_list entry with model: openai/claude-sonnet-4-5, api_base {SITE}/v1, and api_key via os.environ. Or use anthropic/<model> with the root api_base.

anthropic/ or openai/ prefix?
openai/ for the OpenAI-compatible surface (base with /v1). anthropic/ sends native Messages to root + /v1/messages.

LiteLLM keeps appending /v1/messages — how to stop it?
Set LITELLM_ANTHROPIC_DISABLE_URL_SUFFIX=true and give the full path.

Can one LiteLLM endpoint reach multiple models?
Yes — AI Prime Tech is multi-model, so LiteLLM can route to Claude, GPT, and Gemini through the same gateway.

AI Prime Tech is an independent API gateway. It is not affiliated with, endorsed by, or a reseller of Anthropic. Claude and related model names are trademarks of their respective owners.