OpenAI ProviderTools that require user-execution (e.g. apply_patch, shell) are silently dropped / do not work.

TanStack AI version

0.13.0

Framework/Library version

n/a

Describe the bug and the steps to reproduce it

The OpenAI tools shell and apply_patch and are only semi provider tools: OpenAI providers the tool definition, they train the model to use these tools properly - but they're only sometimes run on their servers.

For example, shell supports a local shell mode: { "type": "shell", "environment": { "type": "local" } }]

apply_patch is even more local - it always is supposed to be processed by the API user, not the LLM provider. Calling it a provider-tool is perhaps even a stretch.

Either way, these tool calls are expected to be handled by the user. From OpenAI's docs:

1. Call the Responses API with the apply_patch tool
   • Provide the model with context about available files (or a summary) in your input, or give the model tools for exploring your file system.
   • Enable the tool with tools=[{"type": "apply_patch"}].
2. Let the model return one or more patch operations
   • The Response output includes one or more apply_patch_call objects.
   • Each call describes a single file operation: create, update, or delete.
3. Apply patches in your environment
   • Run a patch harness or script that:
   • Interprets the operation diff for each apply_patch_call.
   • Applies the patch to your working directory or repo.
   • Records whether each patch succeeded and any logs or error messages.
4. Report patch results back to the model
   • Call the Responses API again, either with previous_response_id or by passing back your conversation items into input.
   • Include an apply_patch_call_output event for each call_id, with a status and optional output string.
   • Keep tools=[{"type": "apply_patch"}] so the model can continue editing if needed.

This workflow is currently straight-up impossible with tanstack/ai. Because y'all drop the chunks here: https://github.com/TanStack/ai/blob/main/packages/typescript/ai-openai/src/adapters/text.ts#L307

The chunks (e.g. response.shell_call_command.delta) are just simply never handled.

In general, this function is a bit dangerous in my opinion - it silently drops all unkown chunks, which, you know, can contain important information.

The relevant parts I observed are:

chunk type response.output_item.added
chunk type response.shell_call_command.added
chunk type response.shell_call_command.delta
chunk type response.shell_call_command.delta
chunk type response.shell_call_command.delta
chunk type response.shell_call_command.delta
chunk type response.shell_call_command.delta
chunk type response.shell_call_command.delta
chunk type response.shell_call_command.delta
chunk type response.shell_call_command.delta
chunk type response.shell_call_command.delta
chunk type response.shell_call_command.done
chunk type response.output_item.done

There's no good way at all to recover the chunks (as far as I can tell), and no mechanism to reply with a good output.

In addition, the shell tool clearly should take parameters, but doesn't.

I'm not sure how to best solve this issue myself, because it does touch a couple parts (how'd it be saved in the message history? Would this just be a special tool call? Lots of architecture decisions I'm not comfortable making as a first-time contributor).

But yeah, in the end, e.g. for apply_diff the chunk needs to be surfaced to the user. And there needs to be a way to reply.

Otherwise, ai-sdk is super interesting, and tanstack start is quite nice! But this makes it unfortunately impossible to use for my usecase.

Your Minimal, Reproducible Example - (Sandbox Highly Recommended)

sowwy for not putting it on StackBlitz or sth, but i'm on a train and internet is dire. bun run file.ts should be enough to repro. debug: true surfaces some of the chunks, I think, but I surfaced them by editing node_modules/@tanstack/ai-openai/dist/esm/adapters/text.js - as one does.

import { shellTool } from "@tanstack/ai-openai/tools";
import { chat } from "@tanstack/ai";
import { openaiText } from "@tanstack/ai-openai";

const stream = chat({
  adapter: openaiText("gpt-5.1-codex"),
  tools: [shellTool()],
  messages: [
    {
      role: "user",
      content: "Can you print a simple hello world?",
    },
  ],
  systemPrompts: ["Do what the user tells you to."],
  modelOptions: {
    tool_choice: "required",
    include: ["reasoning.encrypted_content"],
    reasoning: {
      effort: "medium",
      summary: "detailed",
    },
  },
  stream: false,
});

// Observe - nothing is happening.
console.log(await stream);

Do you intend to try to help solve this bug with your own PR?

None

Terms & Code of Conduct

• I agree to follow this project's Code of Conduct
• I understand that if my bug cannot be reliable reproduced in a debuggable environment, it will probably not be fixed and this issue may even be closed.