Local & self-hosted models
Point harnesses at local or self-hosted model endpoints.
There's no dedicated "custom endpoint" field in Proliferate's auth settings. None of the routes in Routes & the picker let you type in an arbitrary base URL. That said, there are two real paths to non-default models today, and it's worth being precise about what each one actually gets you.
Alternative providers through OpenCode
OpenCode fronts a large vendored list of model providers, well over a hundred, including several that aren't Anthropic, OpenAI, or xAI: Groq, Together AI, OpenRouter, LM Studio, and Ollama Cloud among them. Add one from Settings -> Agents -> OpenCode -> Authentication -> Add provider, supply that provider's API key, and OpenCode calls it directly.
OpenCode 'Add provider' search
The searchable Add provider modal open on OpenCode's Authentication tab, showing entries like Groq, OpenRouter, Together AI, LM Studio, and Ollama Cloud, each with its expected environment variable name.
Read the fine print, though: most of these are hosted, API-key services, not a bring-your-own-server option. "Ollama Cloud" and "LM Studio" in that list are the vendors' own hosted inference offerings reached with an API key, not a plain local Ollama or LM Studio daemon running on your machine with no key at all. If your team already has an account with one of these, this is a real and simple way to use it. It isn't the same as pointing Proliferate at a server on your own network.
True local or self-hosted endpoints: use native
If a harness supports pointing itself at a local or self-hosted OpenAI-compatible endpoint through its own configuration, that continues to work in Proliferate as long as you leave that harness on the native route for local runs. Native means Proliferate doesn't touch the harness's environment or config at all; the harness starts up exactly as it would if you'd launched it yourself from a terminal, reading whatever config file or environment variables you've already set up on your machine.
Concretely: if you've configured OpenCode's own config file with a custom provider pointed at a local server, and you haven't turned on the gateway or enabled any API key row for OpenCode in Proliferate, that local provider is what OpenCode will use. Proliferate isn't in the way.
This only works for local runs. A cloud sandbox can't reach a server on your own machine or network, so a harness that depends on a local endpoint can only run there when launched locally, not in the cloud.
What this means in practice
- Need a provider Proliferate doesn't manage directly? Check OpenCode's provider list first; there's a good chance it's already there.
- Need to hit an endpoint that's actually local or self-hosted, with no Proliferate involvement? Configure the harness itself the way you normally would, and make sure it's left on native for local runs.
- Need that same setup for cloud runs? It isn't possible today; a cloud sandbox can't reach a private network.