# Provider connections

Connect the trusted gateway to separately operated inference and preparation roles.

[View canonical HTML](https://pllm.run/cli/reference/gateway/provider-connections/)

Document ID: `pllm.docs.reference.cli.gateway.provider-connections`  
Release: `0.1.0`

Without `--local`, the gateway stays inside the client boundary and connects to provider-role URLs. It does not start those services and it does not make their endpoints OpenAI compatible.

## Credentials

Set `PLLM_INFERENCE_API_KEY` and `PLLM_PREPARATION_API_KEY` in the gateway environment. Keep those role-specific credentials distinct and out of argv, shell history, and checked-in configuration. Public prepared inference needs both URLs; a profile with no preparation role must omit the preparation URL.

## Examples

### Connect to separately operated providers

Set `PLLM_INFERENCE_API_KEY` and `PLLM_PREPARATION_API_KEY` in the environment; credentials should not appear in shell arguments.

```bash
pllm gateway --inference-url https://inference.example.com --preparation-url https://preparation.example.com
```

Start provider roles with [`pllm serve inference`](/cli/reference/serve/inference/) and [`pllm serve preparation`](/cli/reference/serve/preparation/). Both can resolve the same Python, JSON, or YAML experiment target used for local gateway execution and benchmarking.