Shumai

Shumai is an open-source creative collaboration platform built by Yiling-J under the shumaiOne organization, released under the MIT License. It positions itself as an open-source alternative to Frame.io, combining asset management, precise video/image feedback tools, and an AI agent into a single self-hostable workspace. The project reached v0.1.2 as of June 2026 and is written primarily in TypeScript.

What It Is

Shumai is a unified workspace for creative teams to upload files, manage projects, and collaborate with both teammates and an AI agent. Its core differentiator is frame-accurate, coordinate-based annotation directly on video and image assets — enabling precise creative direction without ambiguity. The platform is fully self-hostable via Docker Compose or NPM, and supports S3-compatible storage backends including AWS S3, Cloudflare R2, and MinIO.

Core Feature Set

• Frame-by-Frame Annotations & Comments: Timestamped, drawing-tool-based feedback pinned to specific frames on video and image assets.
• S3-Compatible & Local Storage: Flexible storage backends — local filesystem or any S3-compatible provider.
• Secure Sharing & Collections: Public share links and curated media collections for client and stakeholder collaboration.
• Granular Access Control: Team-level and project-level role-based access controls for workspace permissions.
• Distributed Transcoding via Temporal: Background video transcoding offloaded to a worker pool orchestrated by Temporal.
• Custom Asset Metadata: Dynamic, user-defined metadata fields tailored to production pipelines.

Shumai Agent

The platform includes a built-in AI co-production assistant called Shumai Agent. Key capabilities include:

• Collaborative AI Chat: Context-aware conversation directly within project workspaces.
• Custom Skills & Tools: Extend the agent by registering custom scripts, tools, and automation skills.
• Isolated Sandbox Execution: Agent-submitted scripts run in a sandboxed environment using bubblewrap.
• AI-Powered Metadata Autofill: Automatic tag, description, and metadata generation for new assets using Google Gemini.
• Semantic Search: Vector-embedding-based search for locating assets by visual or conceptual queries.

Deployment Model

Shumai offers three installation paths. Docker Compose is the fastest route — no repository clone required, just a single docker-compose.yaml download. An NPM package (@shumai-one/shumai) supports global installation via npm, pnpm, or bun, with a CLI for daemon management (shumai -d, shumai stop, shumai logs). A source-based development setup is also available for contributors. All paths require PostgreSQL with the pgvector extension; system dependencies include ffmpeg, bubblewrap, socat, and ripgrep on Linux.

Update: v0.1.2

The latest release is v0.1.2, published on June 25, 2026. The repository was created in May 2026 and has seen active development, with the last push on June 25, 2026. The project is tagged with topics including frameio and seedance, signaling its positioning as a Frame.io alternative with AI-assisted video production workflows. The GitHub repository had 118 stars and 4 forks as of the last recorded update.