Skip to content

Web interface

The interface is built with FastAPI + HTMX (src/sbobinator/ui/server.py), default port 8501 (Docker cpu profile: host 8502).

Starting

sbobina ui
sbobina ui --port 9000
start.bat
python scripts\restart_ui.py
Item URL Description
Home / Upload, live queue
Queue & history /jobs Full list, filters, actions
Job detail /jobs/{id} Transcript, summary, downloads, actions
LLM summary /settings/summary API keys, engine status, Qwen download
License /settings/license Terms of use, documentation links, install acknowledgment

Home (/)

File upload, sidebar settings, active queue panel (HTMX every 2 s). After enqueuing, redirect to the new job's /jobs/{id}.

Sidebar: latest 8 jobs with links to the dedicated page.

Queue & history (/jobs)

Table of all jobs with filters, search and quick actions. Click the file name or Open → /jobs/{id}.

Job detail (/jobs/{id})

Dedicated page to review a single job:

  • Breadcrumb "Queue & history / file name"
  • Job in progress: progress panel with HTMX polling (/partials/job/{id}/status), auto-reload on completion
  • Job completed: transcript, summary, TXT/SRT/summary downloads
  • Folder path, Open folder, reprocess, delete

Summary settings (/settings/summary)

  • Cloud provider API keys (DeepSeek, OpenAI, …)
  • Engine availability status
  • Download / path of the local Qwen model
  • System RAM vs minimum threshold (16 GB for local)

Background worker

When the UI starts, start_background_worker() launches a separate process:

python -m sbobinator.cli worker

NeMo does not run in the uvicorn process (avoids lightning.fabric crashes).

PID saved in data/output/jobs/worker.pid.

At worker startup: recover_orphaned_running_jobs() + reconcile_jobs_with_disk().

HTTP API (main)

Method Path Use
POST /enqueue Upload and enqueue
GET /partials/queue HTMX queue fragment
GET /partials/job/{id}/status Single job progress
GET /jobs/{id} Job detail page
POST /api/jobs/{id}/cancel Cancel (optional return_to param)
POST /api/jobs/{id}/delete Delete job and folder
POST /api/jobs/{id}/reprocess New processing
POST /api/jobs/{id}/requeue Requeue
POST /api/jobs/reconcile Sync DB ↔ disk
GET /download/{id}/{kind} Download txt / srt / summary

Tips

  • Only one restart_ui.py at a time — avoid multiple instances on the same port
  • After clean_output.py, reload the page (F5)
  • First job after startup: 1–2 min to load Parakeet into RAM
  • Same file multiple times → multiple folders: use /jobs and the labels with time/summary to tell them apart

See Job queue and history for the full logic.