Configure Channels and Models
Scenario
Create a channel when you need to connect a new model provider, private inference service, or OpenAI-compatible service. Then synchronize and enable models.
Prerequisites
- You have the upstream service URL and credential.
- The gateway node can access the upstream URL.
- The upstream URL complies with the SSRF protection policy.
Create a Channel
- Go to Channel Management.
- Click Create Channel.
- Enter a channel name, for example
openai-prod-east. - Select the channel or protocol type.
- Enter the upstream service URL.
- Enter the upstream credential.
- Configure timeout, priority, weight, and health probe options.
- Click Save.
Synchronize Models
- Select the target channel in Channel Management.
- Click Synchronize Models.
- Wait for the synchronization task to complete.
- Open Model Management or the channel detail page to view synchronized models.
- Enable the models that should be available to users.
Field Description
| Field | Description |
|---|---|
| Channel Name | Identifies the upstream model service in Route Policies and logs. |
| Upstream URL | Address of the upstream model service. It must be reachable from the gateway node. |
| Credential | Authentication information used by the gateway to access the upstream service. |
| Timeout | Maximum wait time for one request. Configure it according to upstream SLA. |
| Priority and Weight | Used by Route Policies to select channels. |
| Health Probe | Used to determine whether the channel can receive traffic. |
Verification
- Run Test Connection or a health probe on the channel detail page.
- Call a target model through
/v1/chat/completions. - Confirm the selected channel in Usage Log.
Notes
Note:
A channel without enabled model rows does not participate in routing. After creating a channel, synchronize models and enable the target models. Otherwise calls may fail with no_available_channel.