v0.19.0

Highlights

• OpenAI Compatible V4: native Chat, Completion, Embedding, and Image models with V4 prompt conversion, provider options, tools, reasoning, detailed usage, metadata, typed HTTP/SSE errors, multipart image editing, and stricter boundary validation.
• Anthropic V4: native language-model, files, and skills surfaces with provider-reference files, container uploads, the advisor_20260301 tool, Claude Sonnet 5, Fable 5, and Opus 4.8 support. Direct xhigh reasoning and provider effort: max remain distinct.
• OpenAI: GPT-5.6 model family, max reasoning effort, Responses reasoning mode/context controls, prompt-cache options and explicit breakpoints, cache-write usage, web_search_call.results, and realtime transport hardening.
• Documentation: README, Starlight provider docs, compile-checked examples, install snippets, changelog, and upstream parity evidence now describe the 0.19.0 surface.

Breaking change

createOpenAI(settings:) and createOpenAIProvider(settings:) now throw when an explicit or environment-provided base URL is empty. Custom provider construction must use try; the default openai facade remains ready to use.

Upstream baseline

Vercel AI SDK commit c8d2726ae045a28142cb46df5e41cdd51d8dcc71, plus the targeted OpenAI latest-main delta at 5b4a299200ec3d061a2cae087d78d13f74d6c90f.

Validation

• Release PR #36 passed GitHub Tests and Vercel checks.
• CI stabilization PR #37 passed GitHub Tests and Vercel checks.
• Final main workflow passed build, all 4158 tests in 473 suites with coverage, coverage report generation, and Codecov upload: https://github.com/teunlao/swift-ai-sdk/actions/runs/29391510132
• Local validation passed swift build, AGENT=1 swift test, AGENT=1 swift test --enable-code-coverage --parallel, examples build, Starlight docs check/build, parity scans, and git diff --check.

Full details are available in CHANGELOG.md.