Skip to Content
DocsLocal Providers

Local Providers

Ziva can discover models from local OpenAI-compatible providers and show them in the model picker. Local providers are useful when you want to run prompts against models on your own machine instead of using hosted AI models.

Currently supported local providers:

  • Ollama
  • LM Studio

How It Works

When the Ziva plugin server starts, it checks each supported local provider at its configured base URL. If the provider is reachable and has models installed, those models appear in Ziva with the provider name shown as the sub-provider.

Local provider models:

  • Run through your locally installed provider
  • Do not use your Ziva AI balance
  • Do not require provider API keys
  • Are listed only when the local provider is running and reachable

Ollama

Install and start Ollama, then make sure you have at least one model installed.

By default, Ziva looks for Ollama at:

http://localhost:11434

Ollama Environment Variables

You can override the Ollama host and port with environment variables:

OLLAMA_HOST=127.0.0.1
OLLAMA_PORT=11434

If only one value is set, Ziva keeps the default for the other value. For example, setting only OLLAMA_PORT=11435 makes Ziva use http://localhost:11435.

LM Studio

Install LM Studio, download a model, and start the local server from LM Studio’s developer or local server panel.

By default, Ziva looks for LM Studio at:

http://localhost:1234

LM Studio Environment Variables

You can override the LM Studio host and port with environment variables:

LMSTUDIO_HOST=127.0.0.1
LMSTUDIO_PORT=1234

If only one value is set, Ziva keeps the default for the other value. For example, setting only LMSTUDIO_HOST=192.168.1.10 makes Ziva use http://192.168.1.10:1234.

Troubleshooting

If local models do not appear in the model picker:

  • Make sure the provider app is running
  • Make sure at least one chat model is downloaded or loaded
  • Check that the configured host and port are reachable from the Ziva plugin server
  • Restart Godot after changing environment variables
  • Refresh the model list from the model picker

If you are connecting to a provider on another machine, make sure that provider is listening on a network-accessible host and your firewall allows the connection.

Last updated on
SkillsOverview