Channel Test

When to use it, which button to click, how to read the result.

When to use

Right after creating a channel: test before enabling.
After a channel has been auto-disabled: running a test re-enables it on success.
When users report "this model isn't working": isolate whether it's the channel.
After editing fields on Add a Channel (Base URL, model list).

Two kinds of test

Single-channel test

Admin → Channels → click a channel → "Test" button.

What it does: sends a minimal request using the channel's actual credentials against one of its allowed models.
Time: typically < 5 s.
Outcome:
- Pass: channel is usable; if it was auto-disabled, it's re-enabled.
- Fail: shows the specific error (401 / 404 / 500 etc.) — diagnose from there.

Batch test

Admin → Channels → "Batch test" at the top of the list.

What it does: tests every enabled channel.
Time: ~1–2 min for 50 channels.
Use: weekly cron to catch channels that silently went bad.
Outcome: full results on the "Test log" page.

Which model to test with

The test model must be in the channel's model allow-list. Safe choices:

OpenAI: gpt-4o-mini (cheap, almost always works)
Anthropic: claude-haiku-4-5
Others: pick the "basic" model from the provider's docs

Error troubleshooting

Error	Meaning	Fix
401 Unauthorized	Wrong / expired credential	Re-issue the API key in the upstream console and update the channel.
404 Not Found	Wrong Base URL	Cross-check against the provider's docs.
400 Bad Request	Format mismatch	Check the Provider type (OpenAI vs Anthropic).
429 Too Many Requests	Upstream rate limit	Lower the concurrency cap or try later.
500 Server Error	Upstream outage	Retry later; pause the channel if it persists.
Timeout	Network	Check if the Base URL is reachable; bump the timeout.