OpenAI Compatible Model Option Gone

[pfeiffee]

At somepoint this option was available:
https://code.visualstudio.com/blogs/2025/10/22/bring-your-own-key
It is no longer available, and all official sources found point to it still being available.
What is the new way to add OpenAI Custom Endpoints?

• Copilot Chat Extension Version: 0.38.2
• VS Code Version: 1.110.1
• OS Version: Arch Linux Latest
• Feature (e.g. agent/edit/ask mode): Language Models -> Add Model
• Selected model (e.g. GPT 4.1, Claude 3.7 Sonnet): NA
• Logs: NA

Steps to Reproduce:

1. Open VSCode with only copilot extension enabled

[DanTup]

On the page you linked it says:

This feature is currently available in VS Code Insiders only.

It seems to be controlled here:

vscode/extensions/copilot/package.json

Lines 1857 to 1859 in 50f36fc

	"vendor": "customoai",
	"when": "productQualityType != 'stable'",
	"displayName": "OpenAI Compatible",

It was removed from stable in b005e33 (PR microsoft/vscode-copilot-chat#1661).

It's not clear to me why this was done, there's no details or related issue on the PR.

Edit:

Some more info here: microsoft/vscode-copilot-release#7518 (comment)

This feature is actually being held to insiders for the forseeable future as a "advanced" tinkerer feature. We leave it up to extension authors to implement specific model providers if they wish.
...
If this feature is important to you, I recommend switching to insiders

It seems odd to me to claim they're being left to extension authors, but add more specific things like Ollama though 🙃

[qisthidev]

Any chance we shipping this today?

matter fact you will need a github account to even add that model lol, then you gonna end up with autocomplete not using that model but using github's copilot features, so you end up sending your data to microsoft

but add more specific things like Ollama though

Yeah I dont think lramos15 has got a good communication with the rest of the team and ends up just pushing the copilot into ditches, all these choices look like they've been done by a person with no critical thinking behind what needs to be implemented before pushing it in a PR, rather than "let's just make this basic feature available and then on top of it we are going to add more"

[ageisen2000]

seems like what was once a free part of copilot is now behind a business paywall

https://code.visualstudio.com/updates/v1_117#_bring-your-own-key-for-copilot-business-and-enterprise

edit: I stand corrected... I have a pro copilot plan and I'm able to do this on vscode insiders, so nvm (for now I guess)

[rossdonald]

You can BYOK in the free version, that's just the enterprise policy.

I think the reason OpenAI compatible is not supported is because it's very complicated to support all the providers properly. They all have slightly different settings, slightly different headers and api requirements. So it's difficult to support fully. I think the vscode team is choosing to focus on other things.

If you want to use OpenAI compatible, you can use the Insiders build or there's extensions like https://github.com/JohnnyZ93/oai-compatible-copilot where you can use your own custom models with Copilot.

[DanTup]

I think the reason OpenAI compatible is not supported is because it's very complicated to support all the providers properly.

I don't think anybody expects them to support all of them, just any of them. There's a spec. If some providers don't follow it, that's on them. And users would probably send PRs if their OpenAI provider didn't work correctly.

But in any case, why lock this to insiders and not just behind an experimental flag? It seems odd to me because it's just pushing people away from Copilot to third party extensions which I'd assume they wouldn't want (because those local model users might also use paid models for bigger tasks and will likely use them in whatever interface also allows their local models)?

[l246804]

Is there any updates?

[qisthidev]

You guys can just installing VSCode Insiders and integrate it with LiteLLM or OpenAI compatible endpoint