Background
llm2 is a Nextcloud External App (ExApp) that runs a local language model and answers AI requests -summarize, free prompt, reformulate, etc. - through Nextcloud's Task Processing framework: a user triggers an action, Nextcloud creates a task, llm2 picks it up, runs the model, and returns the result
Task Processing supports a progress value (0–100%) that a provider can report while it works, and the Assistant renders it as a progress bar. llm2 never reports it, so while the model generates the user sees no real progress - for a long summary the app can look frozen
Proposed change
Make llm2 report an approximate progress while it processes a task, so the Assistant's progress bar advances instead of staying at 0 until the result is suddenly ready. Reporting progress for at least one task type is a good starting point.
"Approximate" is deliberate: a language model generates text token by token and has no inherent notion of percent-complete, so a reasonable estimate — with its limitations understood - is what we are after.
Pointers
- llm2's task handling - its main loop and the per-task processors - lives in this repo.
- The library llm2 uses to talk to Nextcloud is
nc_py_api; look at how a provider reports task
results, and what else a provider can report back while a task runs.
- Nextcloud's developer documentation has a Task Processing section worth reading.
- Under the hood the model runs via llama.cpp (through
llama-cpp-python / LangChain); text
generation is where any real progress signal would come from.
Background
llm2 is a Nextcloud External App (ExApp) that runs a local language model and answers AI requests -summarize, free prompt, reformulate, etc. - through Nextcloud's Task Processing framework: a user triggers an action, Nextcloud creates a task, llm2 picks it up, runs the model, and returns the result
Task Processing supports a progress value (0–100%) that a provider can report while it works, and the Assistant renders it as a progress bar. llm2 never reports it, so while the model generates the user sees no real progress - for a long summary the app can look frozen
Proposed change
Make llm2 report an approximate progress while it processes a task, so the Assistant's progress bar advances instead of staying at 0 until the result is suddenly ready. Reporting progress for at least one task type is a good starting point.
"Approximate" is deliberate: a language model generates text token by token and has no inherent notion of percent-complete, so a reasonable estimate — with its limitations understood - is what we are after.
Pointers
nc_py_api; look at how a provider reports task
results, and what else a provider can report back while a task runs.
llama-cpp-python/ LangChain); textgeneration is where any real progress signal would come from.