Handle parallel tool calls in OpenAI Response API paths - #1420
Merged
iceljc merged 1 commit intoSep 3, 2026
Conversation
The Response API paths kept only the first function call the model asked for, so parallel tool calls were dropped and the model re-asked for the rest on the next turn. The streaming path was worse: each output-item-done update overwrote the previous call, keeping only the last. Collect every FunctionCallResponseItem and populate ToolCalls on the returned RoleDialogModel, matching the Chat API paths. ToolCallId, FunctionName and FunctionArgs still carry the first call, so existing single-call consumers are unaffected. Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Contributor
PR Summary by QodoHandle parallel tool calls in OpenAI Response API paths
AI Description
Diagram
High-Level Assessment
Files changed (1)
|
Contributor
Code Review by Qodo
1. Parallel calls remain unhandled
|
![qodo-code-review[bot]](https://avatars.githubusercontent.com/in/484649?s=60&v=4){kind=small}
| ToolCallId = functionCall.CallId, | ||
| FunctionName = functionCall.FunctionName.NormalizeFunctionName(), | ||
| FunctionArgs = functionCall.FunctionArguments?.ToString(), | ||
| ToolCalls = toolCalls, |
Contributor
There was a problem hiding this comment.
1. Parallel calls remain unhandled 🐞 Bug ≡ Correctness
Although the provider populates ToolCalls, RoutingService.InvokeAgent copies only the legacy first-call fields and invokes one function, so every later requested call is discarded in the standard execution path. The next Responses API request consequently contains only that first call and output, allowing the model to request the omitted calls again.
Agent Prompt
## Issue description
The Responses API provider now returns every requested tool call in `RoleDialogModel.ToolCalls`, but the standard routing pipeline copies and executes only the legacy first-call fields. Update the routing and conversation-history flow to execute each call, preserve each call ID and arguments, and send every corresponding result back to the model while retaining single-call compatibility.
## Issue Context
The same correction must cover non-streaming and streaming Responses API results. Ensure each tool result remains associated with its provider call ID and that completion resumes only after all requested calls have produced results.
## Fix Focus Areas
- src/Plugins/BotSharp.Plugin.OpenAI/Providers/Chat/ChatCompletionProvider.Response.cs[28-49]
- src/Plugins/BotSharp.Plugin.OpenAI/Providers/Chat/ChatCompletionProvider.Response.cs[526-537]
- src/Infrastructure/BotSharp.Core/Routing/RoutingService.InvokeAgent.cs[51-65]
- src/Infrastructure/BotSharp.Core/Routing/RoutingService.InvokeAgent.cs[89-130]
ⓘ Copy this prompt and use it to remediate the issue with your preferred AI generation tools
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
The Response API paths kept only the first function call the model asked for, so parallel tool calls were dropped and the model re-asked for the rest on the next turn. The streaming path was worse: each output-item-done update overwrote the previous call, keeping only the last.
Collect every FunctionCallResponseItem and populate ToolCalls on the returned RoleDialogModel, matching the Chat API paths. ToolCallId, FunctionName and FunctionArgs still carry the first call, so existing single-call consumers are unaffected.