SOP-Hub

A self-hosted clone of Scribe: turn a workflow into a step-by-step visual guide, with AI help from either a local model (Ollama) or a cloud provider (Anthropic Claude, OpenAI, or any OpenAI-compatible endpoint). Runs entirely on your own hardware (e.g. Unraid) via a single Docker container.

Features

• Capture — record a process two ways:
  • Manual: upload screenshots in order and write them up in the editor.
  • Browser extension (extension/): click through the workflow in Chrome and SOP-Hub auto-captures a screenshot + click position for every click, then uploads the whole walkthrough as a new guide in one click.
• Editor — reorder, retitle, and rewrite steps.
• AI assistance, local or cloud, picked per-request:
  • Describe a step from its screenshot (vision).
  • Auto-generate a guide title + summary from its steps.
  • Rewrite step text for clarity/tone.
  • Ask-AI chat that answers questions about a specific guide.
• Publish & share — one click publishes a guide to a public read-only link.
• Export — PDF, HTML, and Markdown, each self-contained with embedded step images.
• Multi-tenant — users, workspaces, and workspace membership.

Quick start

docker run -d \
  --name sop-hub \
  -p 8080:8080 \
  -e JWT_SECRET=$(openssl rand -hex 32) \
  -e PUID=99 -e PGID=100 \
  -v /mnt/user/appdata/sop-hub:/config \
  allornothing/sop-hub:latest

Or with compose (docker-compose.yml in this repo wraps the same image):

cp .env.example .env   # set a real JWT_SECRET at minimum
docker compose up -d --build

Open http://localhost:8080. Local AI is the default (AI_DEFAULT_PROVIDER=ollama) — point OLLAMA_BASE_URL at any Ollama instance on your LAN (it does not run inside this container), or set cloud credentials instead — see AI provider setup below.

Architecture

Everything runs in one container, matching the rest of the AoN-Unraid-Apps collection: Postgres (embedded, data lives under /config/postgres), the Node/Express API, and the built React frontend (served statically by Express on the same port). /config is the single folder to back up — it holds the database and all uploaded step screenshots.

The backend never hard-codes an AI provider: backend/src/services/aiProvider.ts exposes generateText / generateVisionCaption, dispatched by an AiProviderName of ollama | anthropic | openai | custom. Every AI route accepts an optional provider field and otherwise falls back to AI_DEFAULT_PROVIDER. Swapping or adding a provider means editing that one file — feature routes never change.

Dockerfile         multi-stage build: frontend -> backend -> single runtime image
entrypoint.sh       boots embedded Postgres (as PUID:PGID), runs migrations, starts the app
backend/            Node.js/TypeScript + Express API, AI provider abstraction, exports
frontend/           React + Vite SPA
extension/          Manifest V3 Chrome extension -- the "capture" side of the product

Using the browser extension

1. In Chrome, go to chrome://extensions, enable Developer Mode, "Load unpacked", and select the extension/ folder.
2. Click the SOP-Hub Capture icon, log in with your SOP-Hub account and server URL.
3. Click "Start recording" and walk through the process in the browser — every click captures a screenshot.
4. Click "Stop recording" then "Upload to SOP-Hub" — a new guide appears in the app, ready to edit and add AI-generated step descriptions.

AI provider setup

AI_DEFAULT_PROVIDER picks the default; each user's browser can locally override it per-session from the app's Settings page (stored in localStorage, sent as provider on each AI request — no keys ever touch the browser).

Configuration reference

Development (without Docker)

# backend -- needs a Postgres reachable at DATABASE_URL
cd backend && npm install
npm run migrate
npm run dev                      # http://localhost:4000

# frontend -- point it at the backend above
cd frontend && npm install
VITE_API_BASE_URL=http://localhost:4000 npm run dev   # http://localhost:5173

Data model

See backend/src/db/schema.sql — users, workspaces, workspace_members, guides, steps, guide_chat_messages.

Known limitations / next steps

• Step drag-and-drop reordering is up/down buttons, not drag handles yet.
• No image-blur/redaction tool in the editor yet (the schema has a steps.redacted flag ready for it).
• The extension does not yet auto-suggest step titles from the DOM element clicked — it relies on the AI "Describe screenshot" action in the editor instead.
• No role-based permissions beyond workspace membership (all members can edit all guides).

License

MIT — see LICENSE.