Conversations stopping abruptly and randomly

[jmv-gamned]

What happened?

I have an agent that uses Anthropic Sonnet models (the bug occurs more frequently when I choose Sonnet 4.5 rather than 4.0). What I’ve observed, in a random but unfortunately frequent manner, is that the conversation can abruptly cut off in the middle of a discussion. No error message appears, and when checking the agent service logs, it seems that the conversation is indeed considered finished ("final": true), even though the agent hasn’t responded to the user.

Version Information

Running from source : e705b09

Steps to Reproduce

1. Create an agent with Sonnet4.0 or Sonnet4.5 as model
2. Start a conversation
3. Ask for something that triggers the thinking step (in my case : draw me 'something' )

What browsers are you seeing the problem on?

Chrome

Relevant log output

the output of the agent service in my browser :
event: message
data: {"event":"on_message_delta","data":{"id":"step_q8UEqeZJ8ck5_nBRlJ8BM","delta":{"content":[{"index":1,"type":"text","text":"for you!"}]}}}

event: message
data: {"final":true,"conversation":{"_id":"68ef6104eef6c24cfad94bfc","conversationId":"29683829-a175-4d48-bb71-c74c791836ae","user":"67337cc4e5c2029c1adf305c","__v":0,"agent_id":"agent_DANTH5X2olg2_T_DFUlyC","createdAt":"2025-10-15T08:53:24.454Z","endpoint":"agents","expiredAt":null,"files":[],"isArchived":false,"messages":["68ef6104eef6c24cfad94bfb","68ef6109eef6c24cfad94bfe"],"resendFiles":true,"tags":[],"title":"New Chat","updatedAt":"2025-10-15T08:53:29.806Z"},"title":"New Chat","requestMessage":{"messageId":"5f8d9837-6836-4f17-9bd6-600cd11599f7","parentMessageId":"00000000-0000-0000-0000-000000000000","conversationId":"29683829-a175-4d48-bb71-c74c791836ae","sender":"User","text":"Draw me a cute little sheep","isCreatedByUser":true,"tokenCount":13444},"responseMessage":{"messageId":"c0ad5c93-3c2f-49cc-b0c6-503578415784","conversationId":"29683829-a175-4d48-bb71-c74c791836ae","parentMessageId":"5f8d9837-6836-4f17-9bd6-600cd11599f7","isCreatedByUser":false,"isEdited":false,"model":"agent_DANTH5X2olg2_T_DFUlyC","sender":"Your Everyday companion","promptTokens":13,"endpoint":"agents","text":"","content":[{"type":"think","think":"The user wants me to draw a cute little sheep. I should use the NanoBananaCreateImage tool to generate this image. The prompt needs to be in English, which it already is. I should make the prompt descriptive to get a good result.\n\nLet me create a detailed prompt for a cute little sheep."},{"type":"text","text":"I'll generate a cute little sheep for you!"}],"attachments":[]}}

Screenshots

Code of Conduct

• I agree to follow this project's Code of Conduct

[jmv-gamned]

I'm sharing with you my agent's settings :

• Provider: Anthropic
• Model: claude-sonnet-4-5
• Temp: 1.00 (default)
• Top P: 0.70 (default)
• Top K: 5 (default)
• Resend Files: true
• Use prompt caching: true
• Thinking: true
• Thinking budget: 2000
• Websearch: false

[jmv-gamned]

Examples of interrupted conversations that I gathered from several users :

[Munkhu1]

It still happens.