MeltCortex is an AI agent that turns natural language video editing commands into production-ready output. It works by generating MLT XML — the same format Shotcut uses — and renders through melt. No cloud dependency. No opaque pipeline. Just open source, all the way through.
MeltCortex's pipeline is built on production-tested open tools. No proprietary engines. No black boxes. Everything is auditable, self-hostable, and extensible.
An LLM interprets the user's edit command and breaks it into structured edit primitives: trim ranges, filter types, transition styles, aspect ratio targets.
Structured commands are converted into valid MLT XML — the same format Shotcut saves as .mlt files. Clips, filters, timeline, transitions: all encoded in the open XML schema.
The melt CLI (bundled with Shotcut, runs headless) consumes the MLT XML and renders the final video file. No GUI required. Fully automatable.
Every command maps directly to an MLT XML element. MeltCortex's editing model learns the schema so you don't have to.
"Cut from 0:23 to 0:41" — removes a segment, closes the gap
"Fade in from black over 2 seconds" — adds afade/videofade filters
"Make it 9:16 for Reels" — scales and pads to vertical format
"Mute the background music during the voiceover" — adjusts track levels
"Crossfade between clips" — adds compositor transitions
"Add a lower-third title with the guest name" — embeds textoverlay filter
MeltCortex doesn't own your editing pipeline. Every component — the LLM layer, the XML builder, the melt renderer — runs on open tools you can audit, self-host, and modify.
Upload your clips to a self-hosted MeltCortex instance, or run it locally on your machine. Your media never leaves your infrastructure unless you want it to.
MeltCortex is the autonomous video editing agent for creators who want the power of professional pipelines without the professional price tag or the proprietary lock-in.