Using Anthropic Claude, assistant_messages are not streamed, making the streaming endpoint unusable

[seunggs]

Describe the bug

Using Anthropic Claude 3.7 Sonnet, assistant_messages are not streamed when using streaming tokens.

message    {"id":"message-68a02754-a26e-4ac2-b76d-83a0b20f15e9","date":"2025-03-02T18:25:12+00:00","message_type":"reasoning_message","reasoning":""}    
10:25:12.820
message    {"id":"message-68a02754-a26e-4ac2-b76d-83a0b20f15e9","date":"2025-03-02T18:25:12+00:00","message_type":"reasoning_message","reasoning":"\nUser"}    
10:25:12.820
message    {"id":"message-68a02754-a26e-4ac2-b76d-83a0b20f15e9","date":"2025-03-02T18:25:12+00:00","message_type":"reasoning_message","reasoning":" wants a starting draft to work"}    
10:25:14.731
message    {"id":"message-68a02754-a26e-4ac2-b76d-83a0b20f15e9","date":"2025-03-02T18:25:12+00:00","message_type":"reasoning_message","reasoning":" with. I'll create a comprehensive"}    
10:25:14.837
message    {"id":"message-68a02754-a26e-4ac2-b76d-83a0b20f15e9","date":"2025-03-02T18:25:12+00:00","message_type":"reasoning_message","reasoning":" CS teacher persona with patience as a key"}    
10:25:14.941
message    {"id":"message-68a02754-a26e-4ac2-b76d-83a0b20f15e9","date":"2025-03-02T18:25:12+00:00","message_type":"reasoning_message","reasoning":" trait.\n"}    
10:25:15.156
message    {"id":"message-6507791c-9212-420d-ae46-8be227674b30","date":"2025-03-02T18:25:30+00:00","message_type":"tool_return_message","tool_return":"None","status":"success","tool_call_id":"toolu_01C5Ru53pGsy8zKmUCXNs6TC","stdout":null,"stderr":null}    
10:25:30.817
message    {"message_type":"usage_statistics","completion_tokens":391,"prompt_tokens":3460,"total_tokens":3851,"step_count":1}    
10:25:30.960
message    [DONE]

But the message itself is created because once the streaming is over, I can see that assistant_message exists in messages endpoint. It's just not streamed real-time.

Please describe your setup

• How did you install letta?
  Typescript SDK
• Describe your setup
  MacOS (M3 MAX Sequoia), letta server running in docker using the official image