Repository navigation
The Pi agent failing to retrieve the necessary prop information from the server. #117
Description
Error: /props for qwen3.8-flash-next-iq2_xs returned 404
The Pi agent returns the error mentioned above at the end of every prompt, yet it performs inference without issues. The /props command is likely undefined on the server.
If the /props command is defined, we will be able to see the remaining context values, etc.
Activity
hendrikp commentedon Sep 29, 2026
ill take a look at this one and create a PR
Niko1221 commentedon Sep 29, 2026
Fixed by #122 (llama.cpp-style
/models,/propsand/slots), which is merged into the 0.1.23 branch; Pi discovers the model and its context from those. I'll close this when 0.1.23 is out.amonpaike commentedon Sep 29, 2026
I'm testing with Pi right now and it seems to be working well. One question: I have an extension where the model heuristically decides its own "thinking level" , is that actually working, or is it not yet supported at the engine level?
Also, Llama Swap seems to be working now too; with Pi, it's useful for auto-switching, it shifts to smaller models to perform compaction quickly before reloading the heavy model. (a custom compact extension)
For Llama Swap, the config requires
healthCheckTimeout: 300; otherwise, it shuts down before the model has finished loading.Niko1221 commentedon Sep 29, 2026
Fixed in 0.1.23 (released) by #122:
/props,/modelsand/slotsanswer in llama.cpp's format, so Pi finds the model, its context and thinking support. RunSTART-HERE.batto update.serve: add llama.cpp-compatible discovery endpoints
serve: add llama.cpp-compatible discovery endpoints
serve: add llama.cpp-compatible discovery endpoints
serve: add llama.cpp-compatible discovery endpoints