Added embedding API to C++ SDK

[phanindraraja]

Summary

Adds embedding support to the C++ SDK. Also picks up optimizations across the other four SDKs and aligns all five embedding samples to a consistent initialization sequence.

Changes

C++ SDK — new embedding client

Follows the existing chat/audio client pattern: public OpenAIEmbeddingClient with GenerateEmbedding (single) and GenerateEmbeddings (batch), same empty- and whitespace-input rejection as the other SDKs, response parsed into strongly-typed EmbeddingCreateResponse / EmbeddingObject / EmbeddingUsage structs via nlohmann/json. Plumbing (umbrella header, IModel friend, parser overloads, CMake source entry) mirrors OpenAIChatClient one-to-one.

• Tests: 13 unit tests (mocked core, no DLL needed) and one DISABLED_ E2E test matching the shape of DISABLED_DownloadLoadChatUnload.
• Sample: new GenerateEmbeddings example wired into main(); manager initialization now uses a Configuration object so samples can pass core-level settings such as AzureCatalogFilter.
• Build: /EHsc enabled globally so wil's exception-based APIs that core.h relies on resolve against the current vcpkg WIL version.

Optimizations

• C# SDK — materialize the lazy IEnumerable<string> in the batch method once instead of enumerating it three times (Any, per-item validation, and the request builder).
• C# tests — fix a shadowed local in the class setup that could easily mask the static field; remove an in-place genai_config.json mutation helper that left test-data-shared dirty across runs.
• Python SDK — drop the redundant len(inputs) == 0 check (already covered by not inputs).
• Rust SDK — replace the .as_object_mut().map(|m| ...) pattern that discarded its return value with the idiomatic if let Some(obj) = ... plus entry("object").or_insert_with(...).
• Rust docs — drop a stale EmbeddingResponse TOC entry that pointed at a section that was never generated.

Sample alignment across all five SDKs

All embedding samples now run the same seven-step sequence: create manager → discover EPs (prints the table) → download and register EPs → get catalog → get model → download model → load model. The embedding samples previously skipped the EP-discovery step that chat samples show; this brings them into alignment. Model alias is now qwen3-embedding-0.6b everywhere.

Testing

• C++ unit tests: 13/13 pass under ctest --preset x64-debug.
• C++ E2E: runs locally with the Core DLL staged alongside the test binary and --gtest_also_run_disabled_tests.
• All other SDK integration surfaces are unchanged from the user's perspective.

[vercel]

The latest updates on your projects. Learn more about Vercel for GitHub.

Project	Deployment	Actions	Updated (UTC)
foundry-local	Ready	Preview, Comment	Apr 27, 2026 3:30pm

[Copilot]

Pull request overview

Adds an OpenAI-compatible embeddings client to the C++ SDK and aligns surrounding SDK/sample code to support embedding usage consistently across languages.

Changes:

• Introduces OpenAIEmbeddingClient (C++), including JSON parsing for embedding responses and CMake integration.
• Adds C++ unit tests + disabled E2E coverage for embeddings; updates the C++ sample to demonstrate single and batch embeddings.
• Minor cleanups/alignments in Rust/Python/C# embedding clients/tests/docs and updates samples to use the newer embedding alias and to download/register execution providers before running.

Reviewed changes

Copilot reviewed 19 out of 19 changed files in this pull request and generated no comments.

Show a summary per file

File	Description
sdk/rust/src/openai/embedding_client.rs	Simplifies/clarifies response “patching” before deserializing into async_openai types.
sdk/rust/docs/api.md	Removes outdated TOC entry for an embedding response type.
sdk/python/src/openai/embedding_client.py	Simplifies empty-input check (if not inputs).
sdk/cs/test/FoundryLocal.Tests/Utils.cs	Removes the model-config max_length patch helper (no longer used).
sdk/cs/test/FoundryLocal.Tests/EmbeddingClientTests.cs	Removes max_length patch usage; minor variable rename/cleanup.
sdk/cs/src/OpenAI/EmbeddingClient.cs	Avoids multiple enumeration by materializing inputs once and validating the list.
sdk/cpp/test/e2e_test.cpp	Adds disabled E2E test for downloading/loading/unloading and generating embeddings (single + batch).
sdk/cpp/test/client_test.cpp	Adds unit tests validating request format, response parsing, and input/core-error handling for embeddings.
sdk/cpp/src/parser.h	Adds from_json parsers for embedding response structs.
sdk/cpp/src/openai_embedding_client.cpp	Implements the C++ embeddings client over the native embeddings command.
sdk/cpp/sample/main.cpp	Adds a runnable embeddings example (single + batch).
sdk/cpp/include/openai/openai_embedding_client.h	New public C++ header exposing embedding client + response types.
sdk/cpp/include/model.h	Grants embedding client access to IModel::CoreAccess (friend + forward decl).
sdk/cpp/include/foundry_local.h	Exposes embeddings client via umbrella header include.
sdk/cpp/CMakeLists.txt	Enables /EHsc on MSVC and adds the new embeddings source file to the static library.
samples/rust/embeddings/src/main.rs	Updates embedding alias and adds EP discovery + download/register with progress output.
samples/python/embeddings/src/app.py	Updates embedding alias and adds EP discovery + download/register with progress output.
samples/js/embeddings/app.js	Updates embedding alias and adds EP discovery + download/register with progress output.
samples/cs/embeddings/Program.cs	Updates embedding alias and adds EP discovery + download/register with progress output.

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.