Model and Auth Options

Local Docker gives you control. OAuth/subscription login may be cheapest when available. API keys are more predictable. Hosted services are easier. Local models maximize sovereignty but need hardware.

This page is a practical comparison, not a guarantee of provider availability, quotas, or future pricing. ChatGPT/Codex OAuth support can depend on your OpenClaw version, account entitlement, provider policy, weekly quota windows, and re-login requirements.

Option	Best for	Monthly cost structure	Token/API cost	Setup difficulty	Privacy/data control	Laptop resource use	Reliability	Main tradeoff
Local Docker + ChatGPT/Codex OAuth	Users who already have a qualifying ChatGPT/Codex subscription and want a local workspace	Docker is free; subscription cost is paid separately	May be included within subscription limits, but quotas and model access can change	Medium	Strong local file control, but prompts still go to the provider	Moderate for Docker and agent tooling	Good when auth is fresh and routes are supported	Not guaranteed, may have weekly quota limits, and may require re-login
Local Docker + OpenAI API key	Users who want predictable automation and explicit billing	Docker is free; API usage is pay-as-you-go	Billed by API usage under your OpenAI platform account	Medium	Strong local file control, provider sees API requests	Moderate for Docker and agent tooling	Usually predictable for scripts and CI-style workflows	Usage can cost more during long agent sessions
Hosted OpenClaw Launch style service	Users who want less local setup and easier operations	Hosted plan fee, usually separate from any provider usage rules	May be bundled, metered, or bring-your-own-key depending on service	Low	Less local control because runtime is hosted	Low	Often smoother for uptime, updates, and remote access	You trade control and transparency for convenience
Local Docker + Ollama/local model	Users prioritizing sovereignty, offline work, or private experiments	Docker is free; hardware and electricity are the main recurring costs	No external token billing for local inference	High	Highest control when models and data stay local	High, especially for larger models	Depends on your hardware, model size, and Ollama health	Needs capable hardware and may be slower or less capable than hosted frontier models

How To Choose

Choose OAuth if you already pay for ChatGPT/Codex, your OpenClaw version supports the route, and occasional re-login is acceptable.

Choose an OpenAI API key if you need repeatable automation, clearer billing, or fewer surprises around provider routes.

Choose a hosted service if you value fast setup, managed updates, and remote access more than local runtime control.

Choose Ollama/local models if data control matters most and you have hardware that can comfortably run the model you need.

For multi-agent work, model choice can vary by role. Keep the PI Liaison and Scientific Director on the most reliable approved route, then evaluate open-model API candidates for narrower specialist roles. See Model Routing.