runSubagent BYOK Custom Endpoint model naming
- Dominant language
- TypeScript
- Stars
- 193k
- Forks
- 42.4k
- PR merge metrics
- PR metrics pending
Description
- Copilot Chat Extension Version: built in(?)
- VS Code Version: 1.128.0
- OS Version: Windows 11 Enterprise 25H2
- Feature (e.g. agent/edit/ask mode): @RunSubagent available models (BYOK / Custom Endpoint model naming)
- Selected model (e.g. GPT 4.1, Claude 3.7 Sonnet): Kimi-K2.7-code
Hello!
Today I was messing around to see if I could simply tell x AI agent to run the built-in runSubagent MCP with x model, however I noticed something odd.
When adding custom models through the built-in BYOK feature
```
{
"name": "$VendorName",
"vendor": "customendpoint",
```
You cannot set the "vendor"
This causes (for people who model hop a lot) a horridly long list of the same model
the tool gives the result:
ERROR while calling tool: Requested model 'Claude Opus 4.6 ($VendorName)' not found. Available models:
with the list including:
Claude Opus 4.6 (customendpoint),
Claude Sonnet 4.6 (customendpoint),
Claude Opus 4.8 (customendpoint),
Claude Opus 4.5-20251101 (customendpoint),
Claude Opus 4.6 (customendpoint),
Claude Opus 4.6 (customendpoint),
Claude Opus 4.6 (customendpoint),
... and 9 more of the same model
How are AI agents supposed to tell the difference when you cannot set a custom vendor?
It would be nice to either see:
1. models take the name $VendorName (for this example
2. BYOK custom endpoints see support for setting custom vendor - to aid in mcp calls!
Thanks!
Contributor guide
Assessment
This issue has not been assessed yet.