[FIX] Response.json: Body has already been consumed

[ProjectMoon]

Describe the bug

Seems to be related to or caused by #1035. While working through a response, the web app will print out Uncaught (in promise) TypeError: Response.json: Body has already been consumed.. Then it just sits there, and nothing more happens. The spinner just keeps spinning, and the thought process does not advance, nor does it create a response.

To Reproduce

It seems to happen when reading pages via the free Jina API, and those pages taking a long time to be processed. The server logs are something like this (I've removed a bunch of info logs about page loads and context stuffing warnings):

[09:04:23.650689] DEBUG    khoj.processor.tools.online_search: Extracting relevant information from web page at 'https://arxiv.org/abs/2207.11823' took: 31.422 seconds                                 
[09:04:23.653590] INFO     khoj.processor.tools.online_search: Reading web page with Jina at 'https://www.sciencetimes.com/articles/60200/20250107/plutos-largest-moon-formed-kiss-capture-colision-scientists-discover.htm' took: 32.110 seconds                                                           
[09:05:09.848899] DEBUG    khoj.processor.tools.online_search: Extracting relevant information from web page at 'https://www.sciencetimes.com/articles/60200/20250107/plutos-largest-moon-formed-kiss-capture-collision-scientists-discover.htm' took: 46.194 seconds                  
[09:05:31.629136] INFO     khoj.processor.tools.online_search: Reading web page with Jina at 'https://www.astronomy.com/science/astronomers-have-a-new-theory-for-why-uranus-spins-on-its-side/' took: 100.085 seconds                                              

Screenshots

The UI:

Browser console:

Platform

• Server: Self-Hosted Docker
• Client: WebApp
• OS: Linux

If self-hosted

• Server Version [e.g. 1.0.1]: 1.33.2
• Anonymous Mode: Disabled
• Search API: Searxng
• Web scraper: Jina free (no API key set)

Additional context

This happens whether or not research mode is enabled. Other information:

• Using ollama 0.5.4
• Quinn 2.5 14b instruct with a 30k context size
• AMD RX 6800 XT, with 96% of the model loaded into VRAM by ollama and 4% on the CPU.
• It runs fine up until it starts extracting content from webpages with Jina, and then somewhere along the way the API call breaks down in the client.