Agent with Providers and Tools

[ajtiwari07]

Hi,
I am developing a small rudimentary agent that uses redis provider along with a tool which is nothing but a simple function. Below is a small code snippet:

provider = RedisProvider(
       redis_url=redis_url,
       index_name="agent_context",
       prefix="agent_memory",
       application_id=application_id,
       agent_id=agent_id,
       user_id=user_id,
       # Optional: Add vector search for hybrid retrieval
       # redis_vectorizer=your_vectorizer,
       # vector_field_name="vector",
       # vector_algorithm="hnsw",
       # vector_distance_metric="cosine",
   ) 
client.as_agent(
               name="InteractiveAssistant",
               instructions=(
                   "You are a helpful AI assistant. "
                   "You can remember the conversation history and user preferences. "
                   "Be concise, friendly, and provide accurate information. "
                   "If you don't know something, admit it rather than making up information."
               ),
               context_provider=provider,
               tools=[get_weather],
           )

While testing this i encountered that the tool gets correctly invoked the first time only, successive attempts leads to failure. I believe the issue is that framework injects any retrieved text as context during invoking, and Azure AI Foundry (with service-managed threads) doesn't expect external context injection during tool execution flows.
If that is true, then is this a known limitation? Are there any alternatives?

Thanks in advance!

[moonbox3]

Hi @ajtiwari07, what version of the package are you using?

[ajtiwari07]

Hi @moonbox3 thanks for looking into the issue. I am using v1.0.0b260116.

[moonbox3]

Thanks for letting us know, I will try and reproduce the issue.

[moonbox3]

I've created an issue (#3348) from this, so it's easier to track. Please follow along there for any update.