Feature Request: Support OpenAI-compatible base URL override (OPENAI_BASE_URL) for custom/local model routing

[AeonDave]

Describe the feature or problem you'd like to solve

Copilot CLI cannot be routed through OpenAI-compatible proxies (e.g., headroom, LiteLLM, Ollama) because it uses proprietary GitHub endpoints and ignores standard environment variables like OPENAI_BASE_URL. This prevents use with local/self-hosted models and breaks interoperability with existing AI tooling ecosystems that rely on OpenAI-compatible APIs. HTTPS_PROXY does not reliably intercept model traffic, so there is currently no clean way to redirect requests.

Proposed solution

Add support for an environment variable such as OPENAI_BASE_URL or COPILOT_BASE_URL to override the default API endpoint. When set, Copilot CLI should send model requests to the specified base URL using an OpenAI-compatible schema, while optionally preserving GitHub authentication where needed.

Benefits:

• Enables integration with local models (Ollama, LM Studio)
• Allows use of proxy layers (headroom, LiteLLM)
• Improves privacy and offline capabilities
• Aligns Copilot CLI with the broader OpenAI-compatible ecosystem

Example prompts or workflows

1. Route Copilot CLI through a local model:

OPENAI_BASE_URL=http://localhost:11434/v1 -> copilot "explain this function"

2. Use headroom for request routing and logging:

OPENAI_BASE_URL=http://localhost:8000/v1 -> copilot "refactor this code"

3. Switch between providers dynamically:

OPENAI_BASE_URL=https://api.custom-llm.com/v1 -> copilot "generate tests"

4. Offline development workflow:

Run Copilot CLI fully against a local LLM without external API calls

Additional context

• Current behavior bypasses OPENAI_BASE_URL entirely
• HTTPS_PROXY does not intercept model requests (verified: zero traffic observed in proxy tools like headroom)
• Other tools (Aider, Cursor, Claude Code) already support this pattern
• Similar feature requests exist in related Copilot ecosystems but not yet addressed in CLI