abhinavshrivastava950/montara

Montara is a local-first video production system that turns a description of what you want into an actual finished video file. Instead of building for one specific video style, it stores every project as a single editable plan called a Timeline IR, which can then be handed off to different rendering engines depending on what you need and what tools you have installed. Without any paid API keys, Montara can already produce a watchable video using FFmpeg, an open source video processing tool, along with caption cards and local text-to-speech. If you add API keys for cloud services, it can call on additional renderers such as Remotion, Blender, Three.js, Manim, and others, but the project is upfront that those live paid calls only happen when you explicitly turn them on. The README is careful to distinguish between features that are tested today and long-form ambitions, like full length documentaries and movies, that still need more work before they are considered production ready. To get started, you need Node.js, the pnpm package manager, FFmpeg, and Python 3.10 or higher. After installing dependencies, a command called montara doctor checks that everything needed is present before you try to render anything. From there you can run a guided, menu-driven flow to create things like Instagram reels, YouTube shorts, documentaries, or animated explainers, or use non-interactive commands that specify the video type, topic, and length directly. The project also supports exporting to standard editing formats such as EDL, OTIO, and FCPXML, so a video started in Montara can continue being edited in professional tools like Premiere, Resolve, or Final Cut. It is described as agent-ready, meaning humans and AI coding assistants alike can use the same command line tools and skill files to produce videos and leave behind verifiable output. The repository includes public demo videos showing which rendering engines are fully working versus dependent on optional runtimes you may not have installed. The full README is longer than what was shown.