umputun / mpt / 18986698835

88%

push

github

Add GPT-5 support with dual endpoint implementation (#31)

* feat: add GPT-5 support with v1/responses endpoint

Implement native support for GPT-5 models by adding v1/responses endpoint handling alongside existing v1/chat/completions support.

Key changes:
- Add dual endpoint support (v1/responses for GPT-5, v1/chat/completions for other models)
- Auto-detect endpoint based on model name or allow manual forcing via EndpointType
- Add configurable BaseURL field for custom OpenAI-compatible providers
- Eliminate go-openai dependency by implementing direct HTTP client
- Refactor CustomOpenAI to wrap OpenAI provider instead of using external library

Fixes:
- Fix temperature=0 bug where zero temperature was being dropped (changed to *float32 pointer)
- Make API key optional for custom providers that don't require authentication
- Update integration test paths from /chat/completions to /v1/chat/completions

Benefits:
- Unified codebase for all OpenAI-compatible providers
- Full control over request/response handling
- Better testability with direct HTTP mocking
- Smaller binary (removed external dependency)
- Support for both GPT-5 and GPT-4 models

Test coverage: 80.3% (improved from 79.6%)
All tests pass with race detector enabled

* chore: update dependencies and regenerate vendor

Update all outdated dependencies to latest versions:
- anthropic-sdk-go: v1.13.0 => v1.16.0
- mcp-go: v0.41.1 => v0.42.0
- google.golang.org/genai: v1.28.0 => v1.33.0
- google.golang.org/grpc: v1.75.1 => v1.76.0
- golang.org/x/crypto: v0.42.0 => v0.43.0
- golang.org/x/net: v0.44.0 => v0.46.0
- golang.org/x/sys: v0.36.0 => v0.37.0
- golang.org/x/text: v0.29.0 => v0.30.0

Regenerated vendor directory with updated dependencies.
All tests pass successfully.

* feat(custom): add endpoint-type configuration for custom providers

add endpoint-type option to allow custom OpenAI-compatible providers
to specify which API endpoint to use (auto, responses, chat_completions).
this enable... (continued)

236 of 270 new or added lines in 4 files covered. (87.41%)

2286 of 2596 relevant lines covered (88.06%)

34.54 hits per line