Allow subagents.maxConcurrency for BYOK / non usage-based accounts (currently hard-capped at 2)

Look for the subagent concurrency limit logic in the codebase, likely in a settings or configuration module. The issue mentions the setting is ignored for non-usage-based accounts. Start by searching for 'maxConcurrency' and 'subagents' in the source. Check how billing type is determined and where the hard-coded limit of 2 is applied. Testing will require a BYOK setup to verify the change works for non-usage-based accounts.

Summary

subagents.maxConcurrency (added in 1.0.66, "Configure subagent concurrency and depth limits in /settings") is ignored unless the account is on usage-based billing. On other accounts the per-session limit is fixed at 2, and the third task call is rejected immediately (no queue).

Environment

• Copilot runtime 1.0.81 (Windows x64), used through VS Code 1.138 Agent Host
• Account: Copilot Free
• Models: BYOK only (custom OpenAI/Anthropic-compatible gateway); no GitHub-hosted models are used

Repro

1. ~/.copilot/settings.json: "subagents": { "maxConcurrency": 6 }
2. Restart, open a new session, ask the agent to start 4 task subagents in mode: "background" in the same turn.
3. Result: 2 accepted, 2 rejected with:

   Maximum concurrent agent limit of 2 reached. Wait for existing agents to complete before spawning new ones.
   

   telemetry: limit_type=concurrent, limit_value=2, current_value=2, execution_mode=background

The schema explains it: maxConcurrency — "Maximum number of subagents that can run concurrently; applies to usage-based billing users only".

Request

• Let subagents.maxConcurrency apply when the subagents run on BYOK models, since those requests are not billed by GitHub; or
• document which plans get which default, and whether upgrading (e.g. Copilot Pro) unlocks the setting.

Why it matters

Orchestrator + worker patterns (one session plans, several cheap subagents execute) are the recommended way to control cost with BYOK. A fixed cap of 2 per session forces either serialization or several parallel chat sessions for the same task.

Related: microsoft/vscode#336440 (asks for the same knob on the VS Code side).