Offline & network

Run air-gapped, local, or behind a proxy.

#Local models

Point Codey at any OpenAI-compatible endpoint running on your machine — Ollama, LM Studio, llama.cpp servers, or your own. Model discovery runs against the endpoint when you add it.

{
  "provider": {
    "ollama": {
      "name": "Ollama (local)",
      "npm": "@ai-sdk/openai-compatible",
      "options": { "baseURL": "http://localhost:11434/v1" }
    }
  }
}

• Fully local — nothing leaves the machine when the endpoint is on it.
• Zero cost — a local endpoint bills you nothing. It stays grouped under its own provider in the model picker; the pinned Free models group is a separate route, covered in Free models.

#Air-gap

{ "offline": true }

• Zero egress — disables webfetch and remote model calls entirely.

#Proxy

export HTTPS_PROXY=http://proxy.local:8080

• Standard env wins — respects HTTPS_PROXY and NO_PROXY.

#See also

• Configuration
• Models
• Providers