[ENHANCEMENT] Support CoPilot provider natively to optimize credit usage

[wtfzambo]

Problem (one or two sentences)

RooCode currently treats agent tool calls as billable premium requests when using GitHub Copilot via the OpenAI-compatible provider.

This results in unnecessary credit consumption compared to other extensions that correctly distinguish between user-initiated and agent-initiated API calls.

Context (who is affected and when)

This affects users who utilize GitHub Copilot as their primary LLM provider.

When the agent performs multi-step tasks or calls various tools, each internal interaction is being billed against the user's monthly "premium request" quota because the necessary headers to identify agent activity are missing or inconsistent.

Desired behavior (conceptual, not technical)

RooCode should natively support GitHub Copilot to ensure that only direct user interactions count as premium requests.

The extension should correctly identify and flag agent-generated tool calls so they are handled as non-billable background tasks, matching the efficiency found in tools like OpenCode.

Constraints / preferences (optional)

• Billing Accuracy: Ensure compliance with GitHub’s definition of premium requests.

• Header Implementation: Implementation should likely involve specific headers to differentiate user vs. agent traffic (referencing similar fixes in OpenCode issue #8030).

Request checklist

• I've searched existing Issues and Discussions for duplicates
• This describes a specific problem with clear context and impact

Roo Code Task Links (optional)

No response

Acceptance criteria (optional)

No response

Proposed approach (optional)

No response

Trade-offs / risks (optional)

No response

[roomote]

I've analyzed this issue and here's what I found in the codebase:

Current state:

• Roo Code has a vscode-lm provider that uses VS Code's built-in Language Model API (which can interface with Copilot).
• There's also a generic openai provider (OpenAI-compatible) that supports custom base URLs and headers, which users can point at https://api.githubcopilot.com.
• Neither path currently sends the headers needed for GitHub Copilot to distinguish between user-initiated (premium) and agent-initiated (non-premium) requests.

Before proposing a plan, I have some clarifying questions:

1. Which Copilot integration path are you using? Are you using the vscode-lm provider (VS Code Language Model API) or configuring the openai provider with the Copilot API endpoint (https://api.githubcopilot.com)?

2. Header specifics: The referenced OpenCode issue #8030 mentions headers to distinguish user vs. agent traffic. Do you have details on which specific headers the Copilot API expects? From what I can gather, the key headers are likely:

   • Copilot-Integration-Id (identifies the extension)
   • Openai-Intent (e.g., conversation-panel for user requests vs agent for background tasks)
   • Editor-Version / Editor-Plugin-Version

   Can you confirm these or point to more specific documentation?

3. Scope of "native support": Would you prefer:

   • (a) A dedicated copilot provider (new provider type with auto-configuration of Copilot-specific headers, token auth from the Copilot extension, and proper intent differentiation), or
   • (b) Enhancements to the existing vscode-lm and/or openai providers to detect Copilot usage and add appropriate headers automatically?

4. Authentication: When using Copilot's API directly (not via vscode-lm), how are you currently obtaining the OAuth/API token? Are you extracting it from the GitHub Copilot extension's token, or using a personal access token?

Please tag @roomote in your response with the answers, and I'll be happy to propose a detailed implementation plan.

[wtfzambo]

@roomote
Answering the questions:

1. I have been using both OpenAI Compatible provider as well as VSCode-LM (which I just tested). Both of them consume agent tool calls as if they were premium user requests.

2. I am not entirely sure about the header, it could also be something like written here: https://github.com/ericc-ch/copilot-api/pull/85/changes#diff-d128552a12c97629eff117dff90f331b419b4aa87a7dec0c0782ea1748aee3e2R6

3. Because the VScode LM api already exists, then it makes sense to enhance it.

4. I used a token extracted from the extension

[roomote]

Thanks for the detailed answers @wtfzambo! I've seen your comment and will explore enhancing the existing VSCode-LM provider to properly send headers that distinguish agent-initiated requests from user-initiated ones. I'll open a draft PR to share progress shortly.

[roomote]

Opened a draft PR with my attempt at addressing this: #11291

The change adds modelOptions with copilot-integration-id: "roo-code" to VS Code LM API requests when the model vendor is "copilot". This uses the modelOptions property on LanguageModelChatRequestOptions (stable since VS Code 1.96) to help the Copilot backend properly classify agent-initiated requests for billing.

The exact model option key/value may need refinement based on what the Copilot backend actually expects. Testing from users with active Copilot subscriptions to validate the billing impact would be very helpful. Feedback welcome!