Features

• Add Gemini-compatible /v1beta/models endpoints for local gemini-cli usage, including generateContent, streamGenerateContent, and countTokens
• Expose the full upstream model list on the Gemini /v1beta/models surface instead of limiting the listing to a small allowlist
• Add smart fallback routing between /v1/chat/completions and /v1/responses, so requests can still work when a model only supports one of the two OpenAI-compatible endpoints
• Improve OpenAI request conversion compatibility across the two endpoints, including better handling for system instructions, structured output, tool choice, reasoning state, and previous_response_id
• Improve Claude Code native /v1/messages compatibility by removing unsupported passthrough fields before forwarding requests upstream
• Add AmpCode support: chat completions via /amp/v1/* and /api/provider/* route through Copilot API; management routes (/api/*) and login redirects reverse-proxy to ampcode.com

Bug Fixes

• Fix Docker image crash (exec /copilot2api: no such file or directory) caused by dynamically-linked binary in scratch image — add CGO_ENABLED=0 to CI cross-compilation
• Fix Docker multi-arch build: arm64 image was shipping the amd64 binary
• Fix CI triggering redundant runs on tag pushes