OpenMontage
by Calesthio AI Labs
“The first open-source, agentic video production system.” Turns the AI coding assistant you already run into a video production studio — pipelines, tools and production-knowledge files that an agent drives end to end.
See https://www.openmontage.video/ · source at https://github.com/calesthio/OpenMontage
Licence — confirmed, and it matters here
GNU AGPL-3.0 — the full 661-line Affero text, verified in LICENSE and badged in the README. GitHub’s API reports None; the README badge and licence file agree on AGPLv3.
AGPL is the strongest copyleft in common use
Unlike the MIT licences on most tools in this vault, AGPL-3.0 extends copyleft over the network: if you run modified OpenMontage as a hosted service, you must offer users of that service the corresponding source. If you are considering it as a component inside a commercial product or SaaS, this is a real constraint, not a formality. No dual-licence or commercial exception is offered in the repo.
What it is
Rather than a video app with AI features, OpenMontage is a body of pipelines, tools and knowledge that an existing coding agent operates. It ships instruction files for several hosts — CLAUDE.md, CODEX.md, COPILOT.md, CURSOR.md, AGENTS.md, plus an AGENT_GUIDE.md — so the agent you already use becomes the operator.
Python, 517 source files. A mascot (“Monty the Clapper”), a PROMPT_GALLERY.md, and a sponsor-funded model.
Claim check
The project’s headline is “12 production pipelines, 100+ tools, 700+ agent skill and production-knowledge files.” Counted from source:
| Claim | Found | Verdict |
|---|---|---|
| 12 production pipelines | 13 YAML in pipeline_defs/, one being framework-smoke.yaml (a test) | ✅ Accurate — exactly 12 production pipelines |
| 100+ tools | 174 files under tools/, ~183 top-level functions across categories, with a tool_registry.py | ✅ Credible |
| 700+ agent skill and production-knowledge files | skills/ holds 157 files (core 6, creative 36, meta 11, pipelines 103) | ⚠️ Depends on counting — the repo has 1,096 markdown files overall, so 700+ works only if docs and other knowledge markdown are included, not skills/ alone |
The pipelines are real and specific: talking-head, cinematic, documentary-montage, podcast-repurpose, screen-demo, animated-explainer, avatar-spokesperson, character-animation, clip-factory, localization-dub, animation, hybrid.
Superpowers
The genuinely novel move is treating video production as agent-operable knowledge rather than software. Pipelines are YAML, expertise is markdown, and the executor is whatever coding agent you already pay for. That makes the whole system inspectable and forkable in a way a video app never is — you can read exactly what “cinematic” means and change it.
It is also the clearest example in this vault of the pattern that CLAUDE.md-style instruction files have become a distribution format. Shipping five host-specific instruction files is how you make one body of knowledge portable across agents.
Caveats
- AGPL-3.0, as above — the single most important adoption consideration.
- Marketing runs ahead of the repo on the skill-file count, per the table.
- Sponsor-funded with affiliate links in the README (Bloome, Atlas Cloud), which is disclosed but shapes the provider recommendations you will read there.
- Heavy external dependencies — separate
requirements-gpu.txt, ComfyUI and Kling integrations undertools/. Real video generation costs real compute or real API spend; the repo is the orchestration layer, not the models.
Pricing
Free and open source under AGPL-3.0. Actual cost is the generation providers and compute you point it at.
Related
- ComfyUI — integrated under
tools/_comfyuifor local generation - Claude Code · Codex · Cursor — hosts it ships instruction files for
- Multi-Agent Systems — the pipeline/agent pattern underneath