# OpenMontage
OpenMontage is an open source (AGPLv3) agentic video production system. You describe the video you want in plain language, and an AI coding assistant ([[Claude Code]], Cursor, Copilot, Windsurf) handles the whole chain: research, scripting, asset generation, editing, and composition. No manual editing work. It blew up fast: 32k GitHub stars and the #1 trending repository spot.
## How it works
- The agent reads YAML pipeline manifests and Markdown skill files that define production stages and best practices. The same "skills as Markdown" pattern used by [[AI Agents]] harnesses, applied to video
- Tool selection goes through a 7-dimension scoring system (task fit, quality, control, reliability, cost, latency, continuity), and every decision is logged with the alternatives considered. Auditable creative choices instead of a black box
- Composition runs on [[Remotion]] (React) or HyperFrames (HTML/CSS/GSAP); post-production goes through [[FFmpeg]]
- Quality gates before and after render: pre-compose validation, self-review with ffprobe and frame sampling, and a "slideshow-risk" score to catch lifeless output
- Budget controls with cost estimation and per-action approval thresholds
## What stands out
- **12 pipelines**: explainers, animations, talking heads, documentaries, trailers, podcasts, localization
- **52 tools** across video, image, text-to-speech, music, subtitles, and enhancement. Providers include Kling, Runway, [[Veo 3]]-style cloud models, local models (WAN 2.1), [[ElevenLabs]] and Piper for voice, Suno for music
- **Zero API keys needed to start**: ships with free local TTS (Piper) and free stock/archive sources (Pexels, Unsplash, Pixabay, Archive.org, NASA, Wikimedia)
- **Real footage documentaries**: instead of generating everything, it builds a corpus from open archives, retrieves actual motion clips, and edits them into a timeline. A real video, for free
- **Reference-driven creation**: paste a video you like and get differentiated variants instead of starting from a blank page
## Why it matters
The interesting move is architectural: video production expressed as pipelines + skills + tools that any coding agent can drive. The agent harness you already use becomes the video editor. For creators, this collapses the gap between "I have a script idea" and "I have a rendered video" without touching a timeline. I want to test this for YouTube and course content production.
## References
- Repository: https://github.com/calesthio/OpenMontage
## Related
- [[AI Agents]]
- [[Claude Code]]
- [[Remotion]]
- [[FFmpeg]]
- [[ElevenLabs]]
- [[Veo 3]]