A video editor your AI agent drives, you just talk

What it does

gtrk is a TypeScript CLI designed to be called by AI agents (Codex, Claude Code, Cursor, TRAE) rather than by you. You hand it raw footage — a talking-head take, an interview, a script, travel clips — and describe what you want in plain language. The agent orchestrates, gtrk executes: rough cuts, B-roll matching, motion graphics, music, subtitles. Output is a local .gtrk project with everything on editable tracks, viewable and refinable in a companion desktop client, and exportable to CapCut, Premiere, FCPXML, or OTIO.

The interesting bit

The division of labor is the point: the agent handles intent and workflow, the CLI handles deterministic execution, and the human keeps a real editable project rather than a baked MP4. It also chains with Gitruck’s AI Drama Desk, so AI-generated storyboard segments can be laid back onto the timeline as their own track.

Key highlights

• Workflows for oral-cut editing, long-to-short highlight extraction, narration from a script, travel/food storytelling, and picture-in-picture screen recordings
• Uses your own footage library or platform stock, with candidates laid on the timeline for per-segment picking
• Reusable show styles — fonts, colors, visual language — carried across episodes via your own Skills and configs
• Runs on Windows, macOS, and Linux; needs Node.js ≥ 20.6 and FFmpeg
• Real before/after examples in the README (a 1:04 oral take trimmed to 0:48; a 45-second highlight pulled from a 26-minute interview)

Caveats

• Cloud capabilities require an account, API key, and paid quota — some features (cloud-rendered motion graphics, structure export) are billed per use or per minute
• The desktop client for previewing and refining is Windows-only for now
• The client has no built-in AI chat; all prompting happens in your own agent

Verdict

Worth a look if you produce talking-head or narration content and already live inside an agent — it turns “edit this for me” into an actual editable timeline. If you want a fully local, free, one-shot video generator, this isn’t that, and it doesn’t pretend to be.