Prerequisites
Area
Chat / Agent
Problem or Motivation
I think timeouts should be configurable via Odysseus settings, instead of hardcoded like it is now. This would included Deep Research, which has already been proposed, but I believe it should be a universal setting for all models, and configurable on a per model basis.
What this feature enables
By having a long timeout, you can run more models on lower-end hardware, using a swapfile, which dramatically slows down model speed. But by being able to run larger models, the power of Odysseus on these lower-tier platforms is dramatically increased, at the cost of higher time investment.
This means it would be theoretically possible to run the recommended models like DeepSeekv4-flash, or max parameter Qwen3.6 and would also improve the potential experience even on mid-tier to high-tier machines. This also would allow more local model use, and less cloud, making running Odysseus cheaper for everyone.
Proposed Solution
I believe almost all of the logic is in the routes/ subfolder, but there's so many timeout variables it would be cost prohibitive to implement this feature under human power. I think you could get away with only adding this feature to chat, agent, and deep research routes.
Alternatives Considered
I don't believe there is a workaround running through odysseus. It's possible to get some of these harder models to run through ollama or hermes, but it's not anywhere near as powerful.
Prior Art / Related Issues
https://github.com/pewdiepie-archdaemon/odysseus/issues/331
Are you willing to implement this?
No — I am only filing the request
Prerequisites
Area
Chat / Agent
Problem or Motivation
I think timeouts should be configurable via Odysseus settings, instead of hardcoded like it is now. This would included Deep Research, which has already been proposed, but I believe it should be a universal setting for all models, and configurable on a per model basis.
What this feature enables
By having a long timeout, you can run more models on lower-end hardware, using a swapfile, which dramatically slows down model speed. But by being able to run larger models, the power of Odysseus on these lower-tier platforms is dramatically increased, at the cost of higher time investment.
This means it would be theoretically possible to run the recommended models like DeepSeekv4-flash, or max parameter Qwen3.6 and would also improve the potential experience even on mid-tier to high-tier machines. This also would allow more local model use, and less cloud, making running Odysseus cheaper for everyone.
Proposed Solution
I believe almost all of the logic is in the routes/ subfolder, but there's so many timeout variables it would be cost prohibitive to implement this feature under human power. I think you could get away with only adding this feature to chat, agent, and deep research routes.
Alternatives Considered
I don't believe there is a workaround running through odysseus. It's possible to get some of these harder models to run through ollama or hermes, but it's not anywhere near as powerful.
Prior Art / Related Issues
https://github.com/pewdiepie-archdaemon/odysseus/issues/331
Are you willing to implement this?
No — I am only filing the request