Tools on Thread Level not working

[justuswolff-audit]

Hello, I have observed that I can't use tools at the thread level. However, they do work on agent and run level. Let's assume the following script:

# create toolset with code interpreter
toolset = ToolSet()
toolset.add(CodeInterpreterTool())

# create agent, thread, message, and run
agent = agents_client.create_agent(name="Coding Assistant Tool", model="gpt-4.1")
thread = agents_client.threads.create()
message = agents_client.messages.create(thread_id=thread.id, role="user", content="Please execute 2+2 in Python.")
run = agents_client.runs.create_and_process(thread_id=thread.id, agent_id=agent.id)

I could now give tool capabilities in three ways:

1. Agent: agent = agents_client.create_agent(name="Coding Assistant Tool", model="gpt-4.1", toolset=toolset). This works!
2. Thread: thread = agents_client.threads.create(tool_resources=toolset.resources). This does not work!
3. Run: run = agents_client.runs.create_and_process(thread_id=thread.id, agent_id=agent.id, toolset=toolset). This works!

In the first and third case (agent and run) the thread in AI Foundry showed the use of the code interpreter tool, it did not for the second, the thread approach. That is why I am wondering: Am I using a wrong function to set the toolset on the thread? Why can't I pass a toolset but must pass the resources compared to agent and run?

Additional info:

• Tool Overview, describes tool use at the three levels
• Used the Python SDK version = "1.0.0b12", which is the newest as of today

[leestott]

Hi thanks for details and confirming your SDK version

• Thread-level tool configuration is currently non-functional in SDK version 1.0.0b12, even though the method signature accepts tool_resources.
• Unlike agent and run, which actively invoke tools during execution, the thread object is more of a passive container for messages and context. It doesn’t trigger tool execution on its own.
• Internally, the SDK does not propagate tool_resources from the thread to the run, meaning the tools aren’t registered or invoked unless explicitly passed at the agent or run level.

toolset works for agent and run because:

• Agent-level tools are persistent and automatically available to any run initiated by that agent.
• Run-level tools are ephemeral but directly injected into the execution context, ensuring they’re available during that specific run.
• Both these levels actively participate in tool invocation logic, unlike threads.

Recommended approach

Stick to defining tools at the agent or run level for now. If you want thread-specific tool behavior, you’ll need to:

• Pass the toolset explicitly during runs.create_and_process(...)
• Or define it once at the agent level if the tools are meant to be reused across threads

SDK Improvement Suggestion

Please consider opening a feature request at https://github.com/orgs/azure-ai-foundry/discussions/categories/product-feedback-and-ideas You could suggest either:

• Enabling thread-level tool propagation during run creation
• Or clarifying in the docs that tool_resources on threads is currently non-functional