[SDKv2] Various fixes and cleanups from converting the samples to use the V2 API

[skottmckay]

• Fix handling of tool definitions in Responses API when a cached session is found
• Fix delete response to evict session from cache
• Make support for file:// more consistent for image and audio input
• Fix C# setup for ItemQueue
  • it should inherit from Item for consistency
• Fix C# setup for adding Item to Request
  • make ownership transfer optional for consistency
• Fix SDKs to enable getting the overall Response when streaming
• Make getting data from a TensorItem more user friendly

[vercel]

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

Project	Deployment	Actions	Updated (UTC)
foundry-local	Ready	Preview, Comment	Jun 2, 2026 4:51am

[Copilot]

Pull request overview

This PR is a broad set of fixes and cleanups driven by converting the V2 API samples. It reshapes the streaming surface across C#/Python so callers can retrieve the terminal Response (with finish_reason/usage/aggregated items) after iteration completes, makes ItemQueue an Item for C# consistency, makes ownership transfer in Request.AddItem optional, fixes Responses-API tool definitions on cached sessions, evicts cached sessions on DELETE /v1/responses/{id}, normalizes file:// URI handling for image/audio inputs, and adds typed span/array accessors on TensorItem. Several C++ logging tweaks, a field-audit static_assert, and a Smoke.NetFx project removal round it out.

Changes:

• New StreamingResponse wrapper (C#/Python) decouples per-item iteration from terminal Response retrieval, with explicit cancel/dispose/once-only enumeration semantics.
• Responses handler extracts and reapplies tool definitions per request (mirroring chat-completions), and DeleteResponseHandler evicts cached sessions via the new SessionManager::EvictCached.
• file:// parsing centralized in util/file_uri.h (with percent-decoding) and applied to both image and audio inputs; ItemQueue now inherits from Item; TensorItem exposes AsSpan<T>()/ToArray<T>()/ElementCount; Request.AddItem gains a takeOwnership flag.

Reviewed changes

Copilot reviewed 38 out of 38 changed files in this pull request and generated no comments.

Show a summary per file

File	Description
sdk_v2/cpp/src/inferencing/generative/openresponses/response_converter.{h,cc}	Adds ExtractResponsesToolDefinitions (with tool_choice / allowed_tools handling) and field-audit static_asserts.
sdk_v2/cpp/src/service/responses_handler.cc	Applies extracted tools per request and evicts cached session on DELETE.
sdk_v2/cpp/src/inferencing/session/session.h	Adds ClearToolDefinitions().
sdk_v2/cpp/src/inferencing/session/session_manager.{h,cc}	Adds EvictCached(key).
sdk_v2/cpp/src/util/file_uri.h	New centralized file:// parser with percent-decode.
sdk_v2/cpp/src/items/image_item.cc	Uses PathFromFileUri.
sdk_v2/cpp/src/inferencing/generative/audio/audio_session.cc	Uses PathFromFileUri for AudioItem.uri.
sdk_v2/cpp/src/ep_detection/ep_detector.cc	Adds debug logging of detected EP devices.
sdk_v2/cpp/test/* (session_manager_test, response_converter_test, file_uri_test, CMakeLists.txt)	Unit-test coverage for new behaviors.
sdk_v2/cs/src/StreamingResponse.cs	New IAsyncEnumerable+IAsyncDisposable wrapper with FinalResponse task.
sdk_v2/cs/src/Session.cs	ProcessStreamingRequestAsync returns StreamingResponse; producer snapshots cancellation before completing channel.
sdk_v2/cs/src/Request.cs	Adds takeOwnership parameter and null guard.
sdk_v2/cs/src/Items/ItemQueue.cs	Refactored to inherit from Item; resolves inner flItemQueue* once.
sdk_v2/cs/src/Items/Item.cs	FromNative dispatches ItemType.Queue to ItemQueue.
sdk_v2/cs/src/Items/TensorItem.cs	Adds AsSpan<T>, ToArray<T>, ElementCount with type-mapping.
sdk_v2/cs/test/FoundryLocal.Tests/{ChatSessionTests,AudioSessionTests}.cs	Streaming + final-response test coverage.
sdk_v2/cs/test/Smoke.NetFx/*	Project removed.
sdk_v2/python/src/foundry_local_sdk/session.py	New StreamingResponse class and revised process_streaming_request.
sdk_v2/python/src/foundry_local_sdk/openai/live_audio_session.py	Consumes new streaming shape; emits aggregated final transcript.
sdk_v2/python/src/foundry_local_sdk/items.py	MessageItem.is_simple_text()/get_simple_text().
sdk_v2/python/src/foundry_local_sdk/{imodel,foundry_local_manager}.py	Switch ffi.callback signatures to typedef'd names.
sdk_v2/python/src/foundry_local_sdk/init.py	Exports ItemQueue, StreamingResponse.
sdk_v2/python/test/*	Unit/integration test coverage for new APIs.
.github/instructions/cpp-build.instructions.md, memories/repo/*.md	Documentation updates.