A no-spin Shumai review. What the open-source, self-hosted Frame.io alternative does well, where it falls short, the real cost of running it, and who should pick it.
Shumai is one of the more complete open-source, self-hosted Frame.io alternatives available right now. It pairs solid frame-by-frame review with an unusual AI layer — semantic search and Gemini metadata autofill — and it is free under the MIT license. The honest catch: it is early-stage and self-hosted only, with no SaaS option, so it suits technical teams who want data ownership, not solo creators who want zero ops. And it reviews content — it does not generate or publish it.
Most "best Frame.io alternative" lists are stuffed with paid SaaS tools and affiliate links. Shumai is a different kind of entry: an open-source, self-hosted platform you run yourself, with the code on GitHub under the MIT license. It surfaced through a Show HN post in 2026 and has been developed in the open by a small project. That origin shapes everything about who it is for and what trade-offs you accept.
We review this as a content team that lives downstream of tools like this — we generate and publish creative work and care a lot about the review-and-approval step in the middle. So this is not a drive-by listicle blurb. The short version: Shumai is genuinely good at the core job, more ambitious than most open-source review tools thanks to its AI layer, and held back mainly by being young and by the real operational cost of self-hosting.
This review covers what Shumai actually does in 2026, how much it really costs once you count the infrastructure, where it is strong, where it is honestly weak, and who should run it versus reach for a hosted tool instead.
Facts here are reconciled against the public GitHub repository on 2026-06-23. Where the project is young and a detail could change, we kept the claim general rather than precise.
Shumai is an open-source platform for reviewing and managing creative work, positioned as a self-hosted alternative to Frame.io. The core is collaborative review: you upload video and image assets and leave precise feedback with frame-by-frame drawing tools and timestamped comments attached to the footage. Around that sit secure public share links, curated collections for clients, granular team- and project-level role-based access control, customizable asset metadata, and storage on either a local filesystem or any S3-compatible backend (AWS S3, Cloudflare R2, MinIO). Heavy video transcoding is offloaded to a background worker pool orchestrated by Temporal. What sets it apart from a plain review board is the Shumai Agent — an AI layer offering collaborative chat in project workspaces, AI metadata autofill via Google Gemini, semantic search over the library powered by vector embeddings, and custom skills that run in an isolated sandbox. That embeddings layer is why the install requires PostgreSQL with the pgvector extension. It is MIT-licensed, built in TypeScript, installable via Docker compose or the @shumai-one/shumai npm package, and self-hosted only — there is no hosted version to sign up for.
The clearest fit is a technical team, studio, or agency that wants a review-and-approval pipeline on infrastructure it controls — for data-ownership, compliance, or cost reasons — and has the capacity to run Docker, PostgreSQL with pgvector, and S3 storage. Post-production teams handling client video, in-house creative departments that cannot put assets on third-party SaaS, and developers who want to extend a review tool with custom AI skills get the most out of it. It is the wrong tool for a solo creator who wants to sign up and start in two minutes, and the wrong tool for anyone whose actual need is generating or publishing content rather than reviewing it.
| Dimension | Score | Why |
|---|---|---|
| Review & annotation features | 4.2 / 5 | Frame-by-frame drawing and timestamped comments on video and images cover the core Frame.io workflow well. |
| Self-hosting & deployment | 4.0 / 5 | Docker compose gets you running in minutes; the npm path additionally needs Postgres + pgvector and system tools like ffmpeg. |
| Asset management & metadata | 4.0 / 5 | Customizable dynamic metadata fields and collections make it a real media library, not just a comment board. |
| Storage flexibility | 4.5 / 5 | Local filesystem or any S3-compatible backend (AWS S3, Cloudflare R2, MinIO). You own where the bytes live. |
| AI features (Shumai Agent) | 3.8 / 5 | Semantic search, Gemini metadata autofill, and sandboxed custom skills are ambitious for the category — but early. |
| Access control & collaboration | 4.0 / 5 | Team- and project-level RBAC plus secure share links handle client and stakeholder workflows. |
| Maturity & stability | 3.0 / 5 | A young, openly-developed project. Promising and active, but not as battle-tested as mature commercial tools. |
| Cost & value | 4.6 / 5 | Free under MIT with no per-seat fees. The only cost is the infrastructure you run it on — excellent value for a capable team. |
On license cost, Shumai is unbeatable: it is free under MIT, with no seats to buy and no usage meter. Compared to Frame.io and the commercial review tools that charge per seat or per project, a team that already runs its own infrastructure can replace a recurring bill with a one-time setup effort. For the right buyer, that is a large, real saving.
The honest accounting is total cost of ownership, not license price. Running Shumai means a server sized for transcoding, a PostgreSQL instance with the pgvector extension, S3 or local storage you provision and pay for, and the ongoing labor of upgrades, backups, and security patches. For a team with a platform engineer, that overhead is marginal and well worth the control. For a solo creator or a small non-technical shop, the "free" tool can cost more in hours and risk than a $15–$25/month hosted subscription would.
So the pricing verdict is conditional. If you value data ownership and have the technical capacity, Shumai is one of the best value propositions in the review category. If you want to open a browser and start reviewing with zero ops, the free price tag is misleading and a hosted tool is the cheaper choice once your time is counted.
| Use case | Fit | Why |
|---|---|---|
| Self-hosted client video review with frame-accurate notes | Strong | This is the core of the product, and it does it well on infrastructure you control. |
| Studio or agency needing data ownership / compliance over client media | Strong | Assets stay on your own storage backend, which hosted SaaS cannot offer. |
| Organizing a searchable media library with custom metadata | Strong | Custom metadata fields plus AI semantic search make it a real asset-management layer. |
| Solo creator wanting a zero-setup review tool | Weak | No hosted option. You must provision and maintain the whole stack yourself. |
| Mission-critical reliability today | OK | Promising and active, but early-stage — weigh that if downtime is unacceptable. |
| Generating content (captions, clips, images, scripts) | Weak | Out of scope entirely. Shumai reviews and stores content; it does not create it. |
| Publishing and scheduling across social platforms | Weak | No publishing layer. It posts to nowhere — that is a different category of tool. |
| Developer team wanting to extend review with custom AI skills | Strong | The Shumai Agent runs custom skills in a sandbox, and the MIT code is open to extend. |
It is important to be clear that Shumai and Kompozy barely overlap, so this is positioning, not a head-to-head. Shumai is where creative work gets reviewed and approved. Kompozy is where it gets generated and where, once approved, it gets published. If you are evaluating Shumai, you almost certainly already have content to review — the open question is what produces that content and what ships it afterward.
That is the Kompozy slot. It turns one source into 25–35 finished outputs across video, image, text, blog, and newsletter in a governed brand voice, then schedules and posts them across nine platforms. A team that runs Shumai for self-hosted review can sit it directly between Kompozy's generation and Kompozy's publishing: generate the batch, route it through Shumai for client sign-off, then publish the approved cuts. Neither tool is trying to be the other, which is exactly why they pair cleanly rather than compete.
Yes, if you want a free, self-hosted review-and-approval tool and have the technical capacity to run it. It pairs solid frame-by-frame review with an AI layer most open-source tools lack. It is not worth it if you want a zero-setup hosted experience or if your real need is generating or publishing content rather than reviewing it.
The software is free and open-source under the MIT license — no per-seat fees. Because it is self-hosted, your real cost is the infrastructure: a server, PostgreSQL with the pgvector extension, and storage. There is no hosted version offered by the project.
The quickest path is Docker — download the project's docker-compose file, run docker compose up, and open it at localhost:3000. It can also be installed from npm as @shumai-one/shumai, which needs PostgreSQL with pgvector and system tools like ffmpeg. Storage can be local or any S3-compatible backend.
Frame.io is a mature, hosted, commercial review platform. Shumai is open-source and self-hosted — you run it on your own infrastructure under the MIT license. Shumai trades Frame.io's polish and zero-ops convenience for data ownership and no subscription, and adds an AI layer with semantic search and metadata autofill.
Yes — the Shumai Agent offers collaborative AI chat, AI metadata autofill via Google Gemini, semantic search over your library using vector embeddings, and custom skills that run in a sandbox. These are for organizing and searching media, not for generating content.
No. Shumai is a review, collaboration, and asset-management tool. It does not write captions, cut clips, create images, or post to any social platform. For generation and multi-platform publishing you would pair it with an engine like Kompozy.
Skip it if you need a hosted tool with zero setup, if you lack the capacity to run Docker, PostgreSQL, and storage, or if your bottleneck is creating and distributing content rather than reviewing it. The "free" price hides real operational cost for non-technical teams.
For a hosted, mature commercial review tool, Frame.io. For leaner open-source review, tools like Clapshot or MyFrame. For the upstream and downstream problem — generating content and publishing it across platforms — Kompozy, which is a different category that pairs with a review tool rather than replacing one.