Run AoAi thread on different assistants

Question

Good Day,

At Azure Open Ai Assistants, is that possible to run a Thread at different Assistant?

By other words, is the thread coupled with the Assistant?

If so, tokens will differe because of context window?!

Answer 1

Hi Mohamed Hussein

I am able to use same thread_id for multiple assistants in below code.

But I think we have to pass the message from one assistant to another transfer the context and leverage different knowledge resource.

import os
import json
import requests
import time
from openai import AzureOpenAI

# Initialize client
client = AzureOpenAI(
  azure_endpoint="<endpointid>",
  api_key="<endpointkey>",
  api_version="2024-05-01-preview"
)

# Create the first assistant
assistant_1 = client.beta.assistants.create(
  model="gpt-4o",  # replace with model deployment name.
  instructions="Assistant 1 instructions",
  tools=[{"type": "code_interpreter"}],
  tool_resources={"code_interpreter": {"file_ids": ["assistant-FhS84mSrTHuzoNimJPF3s3"]}},
  temperature=1,
  top_p=1
)

# Create the thread
thread = client.beta.threads.create()
print("thread id for assistant is ", thread.id)

# Add a user question to the thread (for Assistant 1)
message = client.beta.threads.messages.create(
  thread_id=thread.id,
  role="user",
  content="Hi from Assistant 1!"  # Replace this with your actual input
)

# Run the thread for Assistant 1
run_1 = client.beta.threads.runs.create(
  thread_id=thread.id,
  assistant_id=assistant_1.id
)

# Wait for the first assistant to complete
while run_1.status in ['queued', 'in_progress', 'cancelling']:
  time.sleep(1)
  run_1 = client.beta.threads.runs.retrieve(
    thread_id=thread.id,
    run_id=run_1.id
  )

# Output results from Assistant 1
if run_1.status == 'completed':
  messages= client.beta.threads.messages.list(thread_id=thread.id)
  print("Assistant 1 Messages:", messages)
elif run_1.status == 'requires_action':
  print("Assistant 1 requires action.")
else:
  print("Assistant 1 status:", run_1.status)

# Create the second assistant
assistant_2 = client.beta.assistants.create(
  model="gpt-4o",  # replace with model deployment name.
  instructions="Assistant 2 instructions",
  tools=[{"type": "file_search"}],
  tool_resources={"file_search": {"vector_store_ids": []}},
  temperature=0.8,
  top_p=1
)


# Add another message to the same thread for Assistant 2
message_2 = client.beta.threads.messages.create(
  thread_id=thread.id,
  role="user",
  content="Hi from Assistant 2!"  # Replace this with your actual input
)

print("thread id used for assistant 2 is", thread.id)
# Run the thread for Assistant 2
run_2 = client.beta.threads.runs.create(
  thread_id=thread.id,
  assistant_id=assistant_2.id
)

# Wait for the second assistant to complete
while run_2.status in ['queued', 'in_progress', 'cancelling']:
  time.sleep(1)
  run_2 = client.beta.threads.runs.retrieve(
    thread_id=thread.id,
    run_id=run_2.id
  )

# Output results from Assistant 2
if run_2.status == 'completed':
  messages_2 = client.beta.threads.messages.list(thread_id=thread.id)
  print("Assistant 2 Messages:", messages_2)
elif run_2.status == 'requires_action':
  print("Assistant 2 requires action.")
else:
  print("Assistant 2 status:", run_2.status)

Thank you