There are 2 updates, 1 comment.
 
 
LLM AI Integration / cid:jira-generated-image-avatar-2f866059-d4ac-42bd-ae69-794236497d66 LLMAI-181 Closed

Chat UI fails with "Unterminated string in JSON" when streaming long responses

 
View issue   ยท   Add comment
 

2 updates

 
cid:jira-generated-image-avatar-18fe1608-d58a-4531-9855-55944784626a Changes by Paul Pantiru on 21/Oct/26 11:52
 
Resolution: Cannot Reproduce
Status: Reopened Closed
 
 

1 comment

 
cid:jira-generated-image-avatar-18fe1608-d58a-4531-9855-55944784626a Paul Pantiru on 21/Oct/26 11:52
 

Closing as Cannot Reproduce. Thanks for the detailed report and for the update that it no longer occurs on 0.10.

We looked for a cause on our side and could not find one: nothing in the streaming path changed between 0.9 and 0.10. The client parser is identical in both versions (it has not changed since 2023), and the server-side streaming code is the same as well, so the fix is not explained by a change in the extension.

The positions reported in the error (12363, 13303, 17438) match the total size of the response bodies you mention (12 to 17 KB). That means the parser was handed one whole body that ended inside a JSON string, i.e. the response was cut short before the final bytes arrived, rather than a JSON message being split across SSE chunks. That points at something between the model and the browser (the reverse proxy or Ollama closing the stream early), which would also explain why it was intermittent and only hit long responses.

If it comes back, please reopen with the server log around the failing request and, if possible, a capture of the raw response body.