feat(sdk/cs): add Responses API client for OpenAI/OpenResponses compat by MaanavD · Pull Request #505 · microsoft/Foundry-Local

MaanavD · 2026-03-10T17:42:01Z

Add OpenAIResponsesClient to the C# SDK v2 with full CRUD support for the Responses API served by Foundry Local's embedded web service.

New files:

src/OpenAI/ResponsesClient.cs: HTTP-based client with SSE streaming
src/OpenAI/ResponsesTypes.cs: Request/response DTOs, items, streaming events
src/OpenAI/ResponsesJsonContext.cs: AOT-compatible source-generated JSON context

Modified files:

src/IModel.cs: GetResponsesClientAsync() on IModel interface
src/ModelVariant.cs: Implementation with web service URL validation
src/Model.cs: Delegation to SelectedVariant
src/FoundryLocalManager.cs: GetResponsesClient() factory method

Key design decisions:

HTTP-based (HttpClient + SSE), not FFI, since no CoreInterop command exists
AOT-compatible: all serialization uses source-generated JsonSerializerContext
IDisposable: HttpClient properly disposed
Follows existing patterns: Utils.CallWithExceptionHandling, ConfigureAwait(false)
Factory on FoundryLocalManager + convenience on IModel
ResponseObject.OutputText convenience property (matches OpenAI Python SDK)
Full CRUD: Create, CreateStreaming, Get, Delete, Cancel, GetInputItems

…tibility Add OpenAIResponsesClient to the C# SDK v2 with full CRUD support for the Responses API served by Foundry Local's embedded web service. New files: - src/OpenAI/ResponsesClient.cs: HTTP-based client with SSE streaming - src/OpenAI/ResponsesTypes.cs: Request/response DTOs, items, streaming events - src/OpenAI/ResponsesJsonContext.cs: AOT-compatible source-generated JSON context Modified files: - src/IModel.cs: GetResponsesClientAsync() on IModel interface - src/ModelVariant.cs: Implementation with web service URL validation - src/Model.cs: Delegation to SelectedVariant - src/FoundryLocalManager.cs: GetResponsesClient() factory method Key design decisions: - HTTP-based (HttpClient + SSE), not FFI, since no CoreInterop command exists - AOT-compatible: all serialization uses source-generated JsonSerializerContext - IDisposable: HttpClient properly disposed - Follows existing patterns: Utils.CallWithExceptionHandling, ConfigureAwait(false) - Factory on FoundryLocalManager + convenience on IModel - ResponseObject.OutputText convenience property (matches OpenAI Python SDK) - Full CRUD: Create, CreateStreaming, Get, Delete, Cancel, GetInputItems

vercel · 2026-03-10T17:42:06Z

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

Project	Deployment	Actions	Updated (UTC)
foundry-local	Ready	Preview, Comment	Mar 10, 2026 5:42pm

Copilot

Pull request overview

Adds a new HTTP-based C# SDK v2 client for Foundry Local’s embedded web service that implements OpenAI/OpenResponses-compatible Responses API operations (including SSE streaming), and wires it into existing model/manager entry points.

Changes:

Introduces OpenAIResponsesClient with CRUD + streaming over HttpClient and source-generated JSON serialization.
Adds Responses API DTOs + custom JSON converters and a dedicated ResponsesJsonContext for AOT/trimming.
Exposes the client via FoundryLocalManager.GetResponsesClient(...) and IModel.GetResponsesClientAsync() (delegated through Model/ModelVariant).

Reviewed changes

Copilot reviewed 7 out of 7 changed files in this pull request and generated 7 comments.

Show a summary per file

File	Description
sdk_v2/cs/src/OpenAI/ResponsesTypes.cs	Adds request/response DTOs, polymorphic items/content, and JSON converters for the Responses API.
sdk_v2/cs/src/OpenAI/ResponsesJsonContext.cs	Adds a dedicated source-generated `JsonSerializerContext` for Responses API types (AOT-friendly).
sdk_v2/cs/src/OpenAI/ResponsesClient.cs	Implements the HTTP/SSE Responses API client and settings container.
sdk_v2/cs/src/IModel.cs	Adds `GetResponsesClientAsync()` to the public model interface.
sdk_v2/cs/src/ModelVariant.cs	Implements `GetResponsesClientAsync()` with loaded-model + web service checks.
sdk_v2/cs/src/Model.cs	Delegates `GetResponsesClientAsync()` to the selected variant.
sdk_v2/cs/src/FoundryLocalManager.cs	Adds a `GetResponsesClient(...)` factory that requires the web service to be running.

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

Copilot · 2026-03-21T02:18:13Z

sdk_v2/cs/src/OpenAI/ResponsesClient.cs

+            throw new FoundryLocalException(
+                "Model must be specified either in the constructor or via GetResponsesClientAsync(modelId).");
+        }


The exception message references GetResponsesClientAsync(modelId), but the manager API added in this PR is FoundryLocalManager.GetResponsesClient(string? modelId = null) and IModel.GetResponsesClientAsync has no modelId parameter. Update the message to reference the correct factory so callers aren’t directed to a non-existent method.

Copilot · 2026-03-21T02:18:14Z

sdk_v2/cs/src/FoundryLocalManager.cs

+    /// <summary>
+    /// Creates an OpenAI Responses API client.
+    /// The web service must be started first via <see cref="StartWebServiceAsync"/>.
+    /// </summary>
+    /// <param name="modelId">Optional default model ID for requests.</param>
+    /// <returns>An <see cref="OpenAIResponsesClient"/> instance.</returns>
+    /// <exception cref="FoundryLocalException">If the web service is not running.</exception>
+    public OpenAIResponsesClient GetResponsesClient(string? modelId = null)
+    {
+        if (Urls == null || Urls.Length == 0)
+        {
+            throw new FoundryLocalException(
+                "Web service is not running. Call StartWebServiceAsync before creating a ResponsesClient.", _logger);
+        }
+
+        return new OpenAIResponsesClient(Urls[0], modelId);
+    }


This PR adds a new public client + factory (GetResponsesClient) but there are existing client-focused tests for chat/audio under sdk_v2/cs/test. Consider adding at least basic coverage for the new factory (e.g., throws when Urls is null/empty, succeeds after StartWebServiceAsync) and one smoke test for create/stream parsing so regressions are caught.

Copilot · 2026-03-21T02:18:14Z

sdk_v2/cs/src/OpenAI/ResponsesClient.cs

+using System.Runtime.CompilerServices;
+using System.Text;
+using System.Text.Json;
+using System.Threading.Channels;


using System.Threading.Channels; is unused in this file. With TreatWarningsAsErrors enabled, this will fail the build (CS8019). Remove the unused using directive.

Suggested change

using System.Threading.Channels;

Copilot · 2026-03-21T02:18:14Z

sdk_v2/cs/src/OpenAI/ResponsesClient.cs

+/// Create via <see cref="FoundryLocalManager.GetResponsesClientAsync"/> or
+/// <see cref="IModel.GetResponsesClientAsync"/>.


The XML doc cref references FoundryLocalManager.GetResponsesClientAsync, but the manager exposes GetResponsesClient(...) (sync). Invalid cref targets typically produce CS1574/CS1584 during doc generation (and warnings are treated as errors). Update the cref (and surrounding wording) to point at the existing API.

Suggested change

/// Create via <see cref="FoundryLocalManager.GetResponsesClientAsync"/> or

/// <see cref="IModel.GetResponsesClientAsync"/>.

/// Create via <see cref="FoundryLocalManager.GetResponsesClient"/> or

/// <see cref="IModel.GetResponsesClient"/>.

Copilot · 2026-03-21T02:18:14Z

sdk_v2/cs/src/OpenAI/ResponsesClient.cs

+        // Cannot use 'using' on response in an async iterator — the response stream's lifetime
+        // is tied to the IAsyncEnumerable consumer. The StreamReader disposal will close the stream.
+#pragma warning disable IDISP001 // Dispose created
+        var response = await _httpClient.SendAsync(httpRequest, HttpCompletionOption.ResponseHeadersRead, ct)
+                                        .ConfigureAwait(false);
+#pragma warning restore IDISP001
+        await EnsureSuccessAsync(response, ct).ConfigureAwait(false);
+
+        var stream = await response.Content.ReadAsStreamAsync(ct).ConfigureAwait(false);
+        using var reader = new StreamReader(stream);


In StreamingImplAsync, the HttpResponseMessage returned by SendAsync is never disposed. Disposing the StreamReader closes the content stream, but the response object itself (and underlying handlers) should still be disposed to avoid resource/socket leaks—especially for long-lived streaming. Use a try/finally in the async iterator to Dispose() the response when enumeration completes or is cancelled.

Copilot · 2026-03-21T02:18:14Z

sdk_v2/cs/src/OpenAI/ResponsesClient.cs

+        [EnumeratorCancellation] CancellationToken ct)
+    {
+        var request = BuildRequest(input, stream: true);
+        options?.Invoke(request);


BuildRequest(..., stream: true) sets Stream = true, but options?.Invoke(request) runs after that and can override request.Stream to false. If that happens, the server may return a non-SSE response while the client still tries to parse SSE, leading to hangs/parse failures. Ensure Stream cannot be overridden for streaming calls (e.g., set request.Stream = true after applying options, or ignore/validate any attempted override).

Suggested change

options?.Invoke(request);

options?.Invoke(request);

// Ensure streaming stays enabled even if options attempts to override it.

request.Stream = true;

Copilot · 2026-03-21T02:18:15Z

sdk_v2/cs/src/IModel.cs

+    /// <summary>
+    /// Get an OpenAI Responses API client.
+    /// Unlike Chat/Audio clients (which use FFI), the Responses API is HTTP-based,
+    /// so the web service must be started first via <see cref="FoundryLocalManager.StartWebServiceAsync"/>.
+    /// </summary>
+    /// <param name="ct">Optional cancellation token.</param>
+    /// <returns>OpenAI.ResponsesClient</returns>
+    Task<OpenAIResponsesClient> GetResponsesClientAsync(CancellationToken? ct = null);


Adding GetResponsesClientAsync(...) to the public IModel interface is a breaking change for any external implementers of IModel (they will no longer compile). If IModel is intended to be publicly implementable, consider using an extension method or a default interface implementation to preserve compatibility.

baijumeswani requested a review from Copilot March 21, 2026 02:14

Copilot started reviewing on behalf of baijumeswani March 21, 2026 02:15 View session

Copilot AI reviewed Mar 21, 2026

View reviewed changes

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

feat(sdk/cs): add Responses API client for OpenAI/OpenResponses compat#505

feat(sdk/cs): add Responses API client for OpenAI/OpenResponses compat#505
MaanavD wants to merge 1 commit intomainfrom
feature/cs-sdk-responses-api

MaanavD commented Mar 10, 2026

Uh oh!

vercel bot commented Mar 10, 2026

Uh oh!

Copilot AI left a comment

Uh oh!

Copilot AI Mar 21, 2026

Uh oh!

Copilot AI Mar 21, 2026

Uh oh!

Copilot AI Mar 21, 2026

Uh oh!

Copilot AI Mar 21, 2026

Uh oh!

Copilot AI Mar 21, 2026

Uh oh!

Copilot AI Mar 21, 2026

Uh oh!

Copilot AI Mar 21, 2026

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

2 participants

		/// Create via <see cref="FoundryLocalManager.GetResponsesClientAsync"/> or
		/// <see cref="IModel.GetResponsesClientAsync"/>.

Conversation

MaanavD commented Mar 10, 2026

Uh oh!

vercel bot commented Mar 10, 2026

Uh oh!

Copilot AI left a comment

Choose a reason for hiding this comment

Pull request overview

Reviewed changes

Uh oh!

Copilot AI Mar 21, 2026

Choose a reason for hiding this comment

Uh oh!

Copilot AI Mar 21, 2026

Choose a reason for hiding this comment

Uh oh!

Copilot AI Mar 21, 2026

Choose a reason for hiding this comment

Uh oh!

Copilot AI Mar 21, 2026

Choose a reason for hiding this comment

Uh oh!

Copilot AI Mar 21, 2026

Choose a reason for hiding this comment

Uh oh!

Copilot AI Mar 21, 2026

Choose a reason for hiding this comment

Uh oh!

Copilot AI Mar 21, 2026

Choose a reason for hiding this comment

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

2 participants