You submits the work, the system generates one video at a time. Controlled by Telegram, visible in a dashboard, and host-agnostic — runs on any bot or standalone. Concurrency controlled to avoid overloading the machine.
Each request becomes a job. The worker picks up one at a time and spawns an autonomous agent session that produces the video from start to finish.
Through Telegram (/mkivideos) or directly. It becomes a job in the queue.
→Persistent (SQLite). FIFO, 1 at a time — doesn't overload.
→Spawns an autonomous Claude session that runs the skill and renders.
→Get notified on Telegram + see it in the dashboard. Submit more whenever you like.
Video rendering saturates the CPU. The queue serializes (concurrency = 1, FIFO) and manages the workflow—you can submit jobs whenever you like.
Ports & adapters: the core doesn't know which bot/DB exists. Runs on openpcbot, openclaw, hermes… or standalone.
Control it with /mkivideos, track it on the dashboard :3141, or run mkivideos run without a bot.
explainer · demo · INEMA courses — and a larger catalog of skills, engines, and AI generators.
A failed job becomes failed and the queue keeps moving. Orphaned jobs from a restart are recovered at boot.
HyperFrames with --gpu --browser-gpu (NVENC) and automatic fallback to CPU.
INEMA courses being turned into videos (1 in-depth video per module). Queue status.
Queue snapshot — 2026-06-08 · 96 jobs total (5 courses)
/mkivideos explicativo "Bayes"
/mkivideos curso <link>
/mkivideos fila
/mkivideos helphttp://localhost:3141/videos
?token=…
→ status, progresso,
cancelar jobsmkivideos add explicativo "X"
mkivideos fila
mkivideos run --port 3141