The Problem
Non-technical creators need to build AI automation workflows without writing code, but existing tools either require technical expertise (n8n, Temporal) or lock users into closed ecosystems. There's a gap for a visual, prompt-driven workflow platform that enables creation, sharing, and monetization without coding barriers.
What This Does
VibeWorkflowPlatform is a NestJS-based backend (apps/api/) for a visual workflow platform called Refly.AI. The codebase organizes functionality into feature modules: auth, canvas, canvas-sync, code-artifact, collab, credit, drive, event, form, knowledge, and copilot. Key infrastructure modules live in apps/api/src/modules/common/: fulltext-search (with elasticsearch, prisma, and token backends), object-storage (fs and minio backends), and vector-search (lancedb, qdrant implementations). Authentication uses OAuth strategies for Google, GitHub, Notion, and Twitter via guards and strategies in apps/api/src/modules/auth/. The system employs Prisma ORM, Redis, and supports both SQL (Prisma) and vector databases (LanceDB, Qdrant) for search and retrieval.
How To Use It
Setup: Install dependencies cd apps/api && npm install
Build the application
npm run build
Start in development mode
npm run start:dev
Configuration: Required environment variables are defined in apps/api/.env.example and apps/api/.env.development. The .env.development file contains what appears to be committed credentials, which is a security concern. Key variables likely include database connection strings, OAuth client secrets, and API keys for vector search/storage services.
Running it: The entry point is apps/api/src/main.ts which bootstrap the NestJS application. Start the development server with npm run start:dev or run via Docker using apps/api/Dockerfile.
Real-World Use
A non-technical creator uses the visual canvas to build an AI automation: they describe a task to the Workflow Copilot, which generates a multi-step workflow on the canvas. The creator connects pre-packaged Agents (nodes) to handle tasks like content generation, data enrichment, and image creation. Once published to the Workflow Marketplace, other users can import and execute the workflow, with the original creator earning credits based on usage. Behind the scenes, the API processes the execution graph, leveraging vector search for semantic document retrieval and object storage for media handling.
Code Health & Issues
High/Security - Possible secrets/credentials committed to apps/api/.env.development - verify and rotate any exposed keys immediately Low/Risk - Dependencies declared without a lockfile in apps/api/package.json - non-reproducible builds; add package-lock.json or pnpm-lock.yaml Medium - 14 test files present but coverage unknown; apps/api/jest.config.ts exists; consider running npm test to validate Medium - CI/CD pipeline configured in .github/workflows/ with 20+ workflows for build, test, and deployment; verify pipeline integrity and secret masking Low - No tsconfig.json at root detected; TypeScript configuration lives within apps/api/
The Bottom Line
This is a functional NestJS backend for a visual AI workflow platform. The codebase is well-structured with modular feature separation and supporting infrastructure for search, storage, and vector databases. However, the lockfile absence and potential secrets in .env.development are concrete risks that should be addressed before considering production use. Best suited for teams wanting to self-host an AI workflow platform rather than individual non-technical creators, given the DevOps overhead of Docker, Prisma, and Redis dependencies.