Aider

Aider treats any OpenAI-compatible endpoint as the openai provider, so pointing it at Inferray is two variables and a model prefix.

Set the variables

export OPENAI_API_BASE="https://api.inferray.com/v1"
export OPENAI_API_KEY="inferray_..."

On Windows use setx OPENAI_API_BASE https://api.inferray.com/v1 and setx OPENAI_API_KEY inferray_..., then open a new shell — setx doesn't affect the session you ran it in.

Run it with the prefix

aider --model openai/claude-opus-4-8

The openai/ prefix is required and easy to miss. It tells Aider to route through the OpenAI-compatible path — the one your base URL points at — rather than trying to infer a provider from the model name. Without it, a model id like claude-opus-4-8 sends Aider looking for Anthropic credentials instead of using your Inferray key.

Everything after the prefix is the model id exactly as Inferray serves it. Any id from the catalogue works, Claude and Gemini included, because they all answer on the same endpoint.

Make it the default

To avoid passing the flag each time, put it in .aider.conf.yml in your home directory or the project root:

model: openai/claude-opus-4-8

Weaker models cost more than they save

Aider edits files by having the model produce exact search-and-replace blocks. A model that formats those loosely burns turns on retries, so the cheapest model is often not the cheapest run. If edits are failing to apply, move up a tier before spending time on configuration.