Is there an open-source alternative to Opus Clip?

Yes. Reelsaw (AGPL-3.0, TypeScript) and OpenShorts (MIT, Python) are both open-source AI clippers you can self-host with Docker Compose. Reelsaw runs with no API key at all and lets you pick any provider per model; OpenShorts keeps a Gemini dependency. See the Opus Clip comparison for a feature-by-feature table.

Does open source mean my video stays private?

Only if every model can run locally. Reelsaw's installer sets up Ollama and a local Whisper so nothing leaves your machine; with cloud models, only audio (for transcription) or transcript text (for selection) is sent to the vendor you chose.

Can I modify Reelsaw and run it for my team?

Yes. The AGPL-3.0 allows internal use and modification freely. If you offer a modified version to the public as a service, you must publish your changes.

Guide

Open-source video clippers: what 'open source' actually gets you

'Open source' on a clipping tool can mean anything from a fully local pipeline to a thin client that still calls one vendor's API. The question that matters is whether you can run every step on your own hardware, with the models you choose. Reelsaw was built to answer yes to both; OpenShorts, the other open-source clipper with traction, answers yes to the first only if you keep a Gemini key.

This page lists what to check and how each project answers.

Four things to check before you choose

Local-only path: can every model run on your hardware, or is one vendor still required for some step? That decides whether 'self-hosted' also means 'private' and 'free'. Reelsaw: yes, by default. OpenShorts: Gemini stays required for layout picking and silent videos.
Provider choice: can you point each model at the vendor you want (or at a local server) without changing code? Reelsaw: every model is a provider:model string, switchable at runtime. OpenShorts: an OpenAI-compatible server for moment picking only.
Licence: MIT lets anyone take the code private, including as a hosted service. AGPL-3.0 requires anyone who runs a modified version for others to publish their changes. Neither is wrong; they protect different people.
Agent surface: an MCP server means Claude Code, Claude Desktop, Cursor or a script can drive the whole pipeline. Check whether it is plan-gated.
Pipeline depth: how moments are chosen, whether filler cuts are verified, what the editor can override, and whether a crashed run resumes.

Reelsaw and OpenShorts, side by side

Reelsaw and OpenShorts, side by side
	Reelsaw	OpenShorts
Fully local, no API keys	Yes, the installer's default: Ollama, local Whisper, on-device alignment and face detection	No: Gemini is still needed for layout picking and silent videos
Language model	Any: OpenAI, Anthropic, Google, Groq, Vercel AI Gateway, Ollama, vLLM, LM Studio or any OpenAI-compatible server; switched at runtime	Gemini by default; any OpenAI-compatible server for moment picking only
Transcription	Any: Deepgram, AssemblyAI, OpenAI, Groq, or a local Whisper server (Speaches, CPU or CUDA)	faster-whisper, optional Parakeet backend; GPU override
GPU	Compose override: faster-whisper on CUDA, Ollama in a container, NVENC encoding; detection models on the CPU	Image build flag: faster-whisper on CUDA, NVENC encoding, detection models on CUDA
Licence	AGPL-3.0, whole repository	MIT core; cloud/ directory under a commercial licence
Stack	TypeScript: Hono, tRPC, Drizzle, Postgres, pg-boss, React	Python: FastAPI, faster-whisper, MediaPipe, YOLOv8; React frontend
Moment selection	Map, pool, rank: chunked at pauses, rubric scored as a median of k samples, pairwise top-10 with swapped positions, optional judge model	Gemini picks 3 to 15 moments per video
Reframing	Face, split, stack, fit per segment; active-speaker following; per-section overrides in the editor	Track, general, split, screencast modes picked per video
Filler removal	Yes, forced-alignment verified, each cut reviewable	Not listed
Editor	Live 9:16 preview, multi-lane timeline, snapping, layout sections, transitions, intros and outros, undo	Dashboard with per-video settings
Agents	MCP server (OAuth optional), operator CLI	MCP server, REST API with webhooks, CLI, agent skill
Beyond clipping	Export zip with platform-ready metadata	Dubbing, AI UGC videos, YouTube studio, direct publishing, hosted cloud
Multi-user	Accounts, admin approval, quotas, usage, runtime model switching	Hosted accounts with API keys
Durability	Every stage is a Postgres job that checkpoints and resumes	Not documented

Read from each project's README on 2026-10-18. Both move quickly; check the repositories for the current state.

Which one to pick

Pick Reelsaw for the core job and control over it: fully local by default with no API key, any provider for any model, selection you can audit, cuts you can review, per-segment framing, an editor, teams, and a pipeline that resumes after a crash.
Pick OpenShorts for breadth around the job: dubbing, UGC generation, publishing, a hosted fallback and a bigger community.
Pick either over a hosted plan if you clip long videos regularly or need the source to stay on your hardware.

Common questions

Is there an open-source alternative to Opus Clip?: Yes. Reelsaw (AGPL-3.0, TypeScript) and OpenShorts (MIT, Python) are both open-source AI clippers you can self-host with Docker Compose. Reelsaw runs with no API key at all and lets you pick any provider per model; OpenShorts keeps a Gemini dependency. See the Opus Clip comparison for a feature-by-feature table.
Does open source mean my video stays private?: Only if every model can run locally. Reelsaw's installer sets up Ollama and a local Whisper so nothing leaves your machine; with cloud models, only audio (for transcription) or transcript text (for selection) is sent to the vendor you chose.
Can I modify Reelsaw and run it for my team?: Yes. The AGPL-3.0 allows internal use and modification freely. If you offer a modified version to the public as a service, you must publish your changes.

Sources

Last updated 2026-10-18.