Model providers

gdc connects to OpenAI-compatible chat-completion endpoints. The backend must support streaming and native tool calls for agent execution.

Supported backends

Backend	Example endpoint
Ollama	http://localhost:11434/v1
llama.cpp server	http://localhost:8000/v1
vLLM	http://localhost:8000/v1
LM Studio	http://localhost:1234/v1
Hosted compatible service	Provider's API base URL

Pointing gdc at a provider

~/.gdc/config.toml

[provider]
base_url = "http://localhost:11434/v1"
model = "qwen3.5:latest"
reasoning = "medium"

Set credentials when the endpoint requires them. CLI flags override environment variables, which override configuration:

gdc --base-url http://localhost:8000/v1 --model loaded-model chat "say hello"

GDC_API_KEY supplies a credential. Its value is hidden in CLI help.

Switching models on the fly

In the terminal, /model opens the provider/model picker; /model <name> selects a model directly. Named endpoints are configured under [provider.profiles]. See configuration.

Choosing a model

Use a model that emits native tool calls on your backend. The local examples use qwen3.5:latest, the project's current tool-use default. A successful gdc chat checks connectivity; gdc print checks tool execution.

Reasoning levels are off, low, medium, high and extra; the default is medium. off omits the provider reasoning field. The terminal picker labels the highest level xhigh.

See troubleshooting for empty replies after tool results.