Yes - three ways, and one of them is a default you never have to configure.
Borrow the model your app already has
Where OpenCues runs inside an app that has its own model, it can route every call through that model instead of asking you for a key. In DeepSeek Harness this is the default: install the plugin and cues and blanks answer on the model your conversations already use, with nothing to sign up for and nothing to paste. The calls are free-standing one-shots - nothing lands in your chat history.
No key ever reaches the page
This matters most on hosts that share their page with other plugins. OpenCues reports which keys exist by name only, hands its runtime a placeholder, and substitutes the real credential locally on the way out - to an allow-list of provider hosts over https only. A plugin sitting beside OpenCues cannot read your key, because the page never holds it.
Reuse a subscription you already pay for
If you have Claude Code or a ChatGPT plan, OpenCues can route through those instead of a metered API key - the same answers, billed to the plan you already have. When no key is configured at all, it falls back to a subscription route automatically rather than dead-ending.
Or stay fully local
Point OpenCues at Ollama and everything runs on your machine, no key and no network egress for the model calls.
The trade, stated plainly
A dedicated fast provider answers in roughly a third of the time of a borrowed host model - measured, same prompt, same machine. For _ blanks, where you are already waiting, the borrowed model feels fine; for passive cues the difference is noticeable. The settings panel puts both numbers in front of you rather than choosing for you, and a host-borrowed setup can usually enable a faster provider inside the host itself and keep every no-key benefit.
Related: Best model for inline AI · Cerebras vs Groq for OpenCues · GitHub repo