FEATURE: BYOM provider registration for generic local inference endpoints (non-Anthropic) #3624

Problem Statement

Copilot CLI supports BYOM providers since v1.0.32+, but the supported providers are limited to Anthropic-specific configurations. There is no configuration path for generic local inference endpoints (e.g., LM Studio, Ollama, llama.cpp) that serve OpenAI-compatible APIs.

This means:

• Local models cannot be registered as BYOM providers for main session tool calls
• Subagent tasks dispatched via runSubagent have no local routing path
• Extensions like VSCode-LMStudio-Bridge cannot expose local models to the Copilot ecosystem

How Other Agents Handle This

Opencode (anomalyco/opencode)

• Provider config system with explicit provider registration
• Supports llama.cpp, Ollama, and generic OpenAI-compatible endpoints
• Subagents inherit session model; provider configuration flows through all dispatch paths

Claude Code (anthropics/claude-code)

• Defaults to cloud but overridable via SG_AGENTIC_MODEL env var
• Model selection is per-request, not session-bound
• Allows explicit model override for subagent tasks

Codex (openai/codex)

• ModelProvider abstraction with routing layer (models_endpoint.rs)
• Supports multiple providers including Amazon Bedrock and local endpoints
• Clean separation between tool execution and model inference

What We Need

1. Generic BYOM provider registration: Allow configuration of arbitrary OpenAI-compatible endpoints as BYOM providers
2. Subagent model inheritance: Ensure subagent tasks respect the session model when it is a local endpoint
3. Provider priority/fallback: Support tiered routing (local first, cloud fallback with warning)

Security Implications

• Data leakage: Workspace context transmitted to cloud without user consent when local models are available
• Cost opacity: Untracked cloud API usage from subagent tasks
• Trust erosion: Silent cloud routing violates local-only expectations
• Compliance risk: Sensitive code may be processed by third-party inference services

Related Issues

• Task tool silently downgrades subagent model to session model via multiplier guard (frontmatter and explicit model override both ignored) #3565 — multiplier guard silently downgrades subagent model
• Feature request: local web interface (like opencode web) #3301 — local model feature request
• ian-morgan99/VSCode-LMStudio-Bridge#337 — bridge-level gap analysis
• ian-morgan99/VSCode-LMStudio-Bridge#336 — zero BYOM integration in bridge

Acceptance Criteria

1. Generic OpenAI-compatible endpoints can be registered as BYOM providers
2. Subagent tasks respect session model when it is a local endpoint
3. Provider priority/fallback configuration is supported
4. Cost/privacy warnings display when cloud fallback triggers
5. Documentation updated with routing behavior and configuration options