Channel Management
Channels are the core configuration unit for connecting AI providers
Channels are the core configuration unit for connecting AI providers. Each channel corresponds to one provider API key. Log in with an admin account, then click "Channels" in the left sidebar or visit /channels directly.
The channel list shows all configured AI provider channels with their name, type, status (green = active / red = disabled), response time, and used quota.
Add a Channel
-
Click the "Add Channel" button in the top-right of the channel list page — a configuration dialog appears
-
Select the provider type (e.g. OpenAI, Claude, Gemini)
-
Enter the channel name and API key
-
Check the models this channel supports in the model list, or click "Fill Default Models" to auto-populate
-
Expand the advanced config section as needed and fill in optional fields:
| Option | Description |
|---|---|
| Base URL | Custom endpoint URL for proxies or self-hosted deployments |
| Priority | Higher value = higher selection priority, default 0 |
| Weight | Random weight among same-priority channels, default 0 |
| Model Mapping | Map user-requested model names to actual model names (JSON) |
| Parameter Override | Force-override certain request parameters (JSON) |
| Auto Disable | When enabled, automatically disables the channel after consecutive failures |
OpenAI Alpha Search
POST /v1/alpha/search is a standalone Web Search relay. It selects only Codex channels or Advanced Custom channels with an explicit native Alpha Search pass-through route. A regular OpenAI channel is not selected merely because it exposes the same model name.
Codex Channel
Choose the Codex channel type and configure its models, credentials, and Base URL. A client request to /v1/alpha/search uses the native Codex path /backend-api/codex/alpha/search. Model mapping and Param Override still apply.
Advanced Custom Channel
Add a route under Advanced Custom Routes:
| Option | Value |
|---|---|
| Incoming path | /v1/alpha/search |
| Upstream path | The upstream Alpha Search path or a complete URL |
| Converter | Native forwarding (none) |
| Auth | Default Bearer, no authentication, custom Header, or Query according to the upstream requirements |
Also add the supported models to the channel model list, then configure model mapping and Param Override as needed. In channel testing, select the OpenAI Alpha Search endpoint to verify routing, authentication, and the upstream response.
Alpha Search preserves unknown fields from the original JSON body, replacing only the mapped model and applying explicitly configured Param Override operations. See OpenAI Alpha Search for client examples and request rules.
Alpha Search adds one web_search_preview tool-call charge to the model reservation. Before production use, verify the default or model-prefix tool price under /system-settings/billing/model-pricing, together with group ratios and tiered billing. Tool usage is recorded only after an upstream success; failed requests follow the existing refund path.
Video Task Billing (DoubaoVideo)
This setting is available only for DoubaoVideo channels. Edit a DoubaoVideo channel, expand advanced settings, and open Video Task Billing to control Completion delta settlement.
- It is enabled by default. The switch state is captured when a task is submitted. When the task completes, the system can use upstream
usage.total_tokensto apply a delta charge or refund against the reserved quota. - Disable it when upstream usage is unreliable, channel pricing has not been calibrated, or submitted tasks must not be adjusted again after completion. When disabled, the task keeps its submission-time reserved or fixed billing result and no automatic completion-time charge or refund is applied.
- The switch affects DoubaoVideo only; other video channels ignore
disable_task_delta_settlement. It is stored in the channelsettingsJSON and requires no database migration.
New tasks retain the submission-time setting. Older unfinished tasks without a snapshot fall back to the channel's current setting. Maintain ModelRatio and task-billing configuration against actual upstream cost regardless of this switch.
- Click "Submit" — the new channel appears in the list
Test a Channel
- Find the target channel in the list and click the "Test" button in the action column on the right
- Wait for the test request to complete — a dialog shows the response time and success/failure status
Lower response time means faster performance. You can also click "Test All Channels" at the top of the list for one-click batch testing.
Batch Operations
-
Check the checkboxes on the left side of multiple channel rows
-
A batch operation toolbar appears at the top of the page
-
Click the corresponding button to perform the batch operation:
- Batch Enable: Set selected channels to active
- Batch Disable: Set selected channels to disabled
- Batch Tag: Assign a tag to selected channels for categorized management
Multi-Key Mode
Multi-Key mode lets a single channel use multiple API keys with automatic polling. A failed key is skipped automatically and re-enabled when it recovers.
-
Click the "Edit" button on the right side of the target channel row
-
Find the "Multi-Key Management" section in the edit dialog
-
Click "Add Key" to enter multiple API keys one by one
-
Select a polling mode:
| Polling Mode | Description |
|---|---|
| Round Robin | Use each key in sequence |
| Weighted Random | Select keys randomly by weight |
- Click "Save" to complete the configuration
How is this guide?