Enterprise BYOK: allow gateway-fronted Azure AI Foundry / Anthropic-compatible endpoints for regulated zero-trust architectures #198472
Replies: 2 comments
-
|
💬 Your Product Feedback Has Been Submitted 🎉 Thank you for taking the time to share your insights with us! Your feedback is invaluable as we build a better GitHub experience for all our users. Here's what you can expect moving forward ⏩
Where to look to see what's shipping 👀
What you can do in the meantime 💻
As a member of the GitHub community, your participation is essential. While we can't promise that every suggestion will be implemented, we want to emphasize that your feedback is instrumental in guiding our decisions and priorities. Thank you once again for your contribution to making GitHub even better! We're grateful for your ongoing support and collaboration in shaping the future of our platform. ⭐ |
Beta Was this translation helpful? Give feedback.
-
|
For info : github/roadmap#1027 |
Beta Was this translation helpful? Give feedback.
Uh oh!
There was an error while loading. Please reload this page.
-
🏷️ Discussion Type
Product Feedback
💬 Feature/Topic Area
Copilot in GitHub
Body
Hi GitHub Copilot team,
I would like to raise an enterprise BYOK feature request for GitHub Copilot custom models.
GitHub Support has already confirmed that this is not a troubleshooting or configuration issue, but a product limitation / feature gap. They suggested opening a public discussion so other enterprise customers can signal demand.
Summary
GitHub Copilot BYOK currently blocks a valid regulated architecture where Azure AI Foundry is fronted by an enterprise API gateway.
The backend model endpoint works correctly. The blocker is that the Copilot BYOK configuration flow validates the Microsoft Foundry endpoint hostname before sending any request, and rejects gateway-fronted URLs that are intentionally not
*.services.ai.azure.comor*.openai.azure.com.For regulated organizations, this prevents adoption of Copilot custom models with private Azure AI Foundry deployments.
Enterprise architecture
We are operating in a regulated financial services environment with a hardened Azure Landing Zone and a central Azure AI Foundry account.
The Foundry account is configured with:
public_network_access_enabled = falselocal_auth_enabled = falseThe intended architecture is:
Consumers never receive a Foundry key or direct Foundry endpoint access.
What works today
APIM exposes an Anthropic-compatible endpoint:
Direct protocol tests with Postman and curl are successful.
The APIM endpoint returns clean HTTP 200 responses with Claude Opus through Azure AI Foundry.
So the AI plumbing, authentication, APIM policy, backend routing, and response format are all working.
Current blocker
The Microsoft Foundry BYOK provider rejects the APIM URL during wizard validation before any request is sent.
This has been confirmed because APIM receives zero access log entries during the Copilot BYOK validation attempt.
The issue is therefore not connectivity, authentication, APIM policy, or model compatibility.
The issue is that the BYOK configuration UI enforces a hostname pattern such as:
This validation blocks enterprise gateway-fronted endpoints even when they are fully compatible with the expected provider API.
The Anthropic and OpenAI BYOK providers do not currently provide a custom base URL field either, so the same architecture is blocked through those providers as well.
Requested product changes
There are two possible ways to solve this.
Preferred option
In the Microsoft Foundry BYOK provider, add an explicit opt-in option to allow a gateway/proxy endpoint.
For example:
When this option is enabled, Copilot should skip strict hostname validation and validate the endpoint by making the actual protocol call instead.
Alternative option
Expose a
Base URLfield for Anthropic and OpenAI BYOK providers, allowing operators to configure any compatible endpoint.For example, an Anthropic-compatible Messages API endpoint could point to:
This would allow enterprise gateways, private proxies, APIM, internal model gateways, or regulated model brokers to work without requiring Copilot to support every gateway implementation explicitly.
Why this matters
In regulated enterprise environments, direct public exposure of Azure AI Foundry endpoints is often not acceptable.
An enterprise API gateway is not an optional convenience. It is part of the control plane.
It provides:
Without support for gateway-fronted BYOK endpoints, regulated organizations are forced into bad options:
Expected behavior
Copilot BYOK should support enterprise-controlled endpoints when they are protocol-compatible with the selected provider.
A gateway-fronted Microsoft Foundry / Anthropic-compatible endpoint should be accepted if:
Actual behavior
The endpoint is rejected before any outbound request is made because the hostname does not match the expected Microsoft Foundry / Azure OpenAI domain pattern.
No request reaches APIM during validation, which proves this is a preflight hostname validation issue rather than an API/runtime failure.
Why this is broader than one customer
This pattern is common in regulated industries:
Many enterprise customers cannot let every AI-consuming tool connect directly to backend model providers. They need a central broker for governance, audit, security, and cost management.
Supporting custom/gateway-fronted BYOK endpoints would make GitHub Copilot custom models much more usable in enterprise zero-trust environments.
Requested feedback from GitHub
Could the Copilot product team consider adding either:
Base URLfield for Anthropic/OpenAI/OpenAI-compatible BYOK providers in the enterprise BYOK configuration flow?I would be happy to provide architecture diagrams, APIM policy examples, sanitized traces, and request/response samples if helpful.
Thanks!
Beta Was this translation helpful? Give feedback.
All reactions