Shaitrish/37studio

37Studio is best understood as a self-hosted replacement for the video-editing and social-scheduling layer of a marketing stack, not a hands-off content machine. It suits people who already have brand material and platform developer apps, and who are willing to run Docker.

The problem

A small brand or solo operator wants a short branded video out on Instagram, TikTok, YouTube and X this week, then wants to answer the replies. Today that means cutting in an editor, exporting per-platform files, writing separate captions, uploading each one by hand, and checking inboxes one by one. 37Studio moves the assembly into a chat with an editing agent and puts the finished render into a posting queue that reaches nine platforms, with an inbox loop for Messenger and Instagram messages. The output is a vertical composition plus per-platform captions, rather than a stack of manual exports.

What's different

The mechanism is a Go server that wraps claude -p in a project folder. The agent gets the brand's knowledge base, design system and asset library, probes and transcribes footage, cuts with ffmpeg, writes an animated HyperFrames composition, then renders stills to inspect its own frames before making changes. You can also edit on a timeline by hand. The practical benefit is an iterative loop in plain language rather than remembering ffmpeg syntax, and a hard per-turn budget caps Claude spend.

The trade-off is that quality and cost depend heavily on the AI provider. The default tier is Opus, with Haiku, Sonnet and Codex alternatives documented, and each run consumes plan or API budget. Brand files and source folders are sent to Anthropic as part of the agent's work, and the launcher and cloud file set bypass agent permissions, letting the agent run any command inside the container; running from source defaults to a safer allowlist. Publishing also depends on your own Meta, Google, TikTok, LinkedIn, X or Pinterest developer apps, which is real setup work. See the setup guide.

Who should try it

Try it if you run content for one or a few brands, already hold the platform developer credentials or are willing to create them, and want the editing and scheduling loop on a machine you control. It fits a team that can leave a 16 GB Linux VPS running so the post queue and auto-replies continue while laptops are closed. It is a poor fit if you want an upload-and-forget service, if you do not want agent-read brand material leaving your machine, or if a Claude plan or API key is not something you want to pay for. A ChatGPT plan plus Codex sign-in is a documented alternative to an Anthropic key, and Gemini and ElevenLabs are optional extras; that changes the cost picture but not the Docker requirement.

Requirements and limitations

You need Docker Desktop or Docker Engine, 16 GB of RAM and about 20 GB of free disk, plus a Claude plan or Anthropic API key; the cloud path recommends a 16 GB VPS with 4+ vCPUs and 50 GB of disk because rendering and transcription run on CPU. Platform connections are not turnkey: each service needs your own developer app and OAuth client, so onboarding is a sequence of console tasks. The Linux /data permission issue documented in the FAQ is a reminder that ongoing ownership of the data folder matters.

Ongoing upkeep includes token refresh, platform API changes and provider model menus, since the agent lists current models per connected provider. Anthropic's terms govern using a subscription token in another app; the README calls an API key the fully supported route. WhatsApp inboxing needs a Cloudflare Tunnel or your own relay, which adds a network component to look after.

HotRepo's take

HotRepo's only comparable record is a #9 placement on the NEW-REPOSITORY board (新项目榜, not the overall Trending board) on 2026-10-25, with 63 total stars two days after creation. That is a snapshot, not a growth series, and persistence cannot yet be judged. Before committing, check whether each platform you need has a working developer-app path in the platform table and confirm the agent's file access matches what you are comfortable sending to a model provider.