ModelEngine-Group/app-platform

AppPlatform – Low‑code platform for building AI‑driven applications

What it is – A Java‑backed, React‑front‑end platform that lets product teams and engineers assemble, test and deploy AI‑enabled services (chatbots, RAG pipelines, autonomous agents, etc.) without writing a lot of boiler‑plate code. It is built on the open‑source FIT framework and the Waterflow workflow engine, and uses the ELSA graphical editor for visual flow design.

Key components

• Backend – Java 17, plugin‑based architecture (FIT). Provides an application manager for creating, debugging and running AI apps, and a function‑node system that lets you plug in custom operators written in Java, Python or other languages. Execution is orchestrated by Waterflow.
• Frontend – React functional components. Offers a visual low‑code canvas (ELSA), an app marketplace, a JSON‑Schema‑driven smart‑form builder, and a plugin manager for uploading custom extensions.
• Low‑code UI – Users drag‑and‑drop nodes, connect them, and configure model endpoints via a web UI; no deep coding required.
• Multi‑model orchestration – Different LLMs can be chained together inside a single flow, enabling RAG, tool‑calling, or agent‑style interactions.
• Store & reuse – Finished apps, functions, and “tools” are stored as reusable templates that can be shared across projects.

How to get started

1. Prerequisites – Docker & Docker‑Compose, at least 2 CPU cores, 4 GB RAM, PostgreSQL ≥ 14 (for production). The repo ships a docker/.env.example you copy to .env and fill in model URL/API‑key, then run bash docker/deploy.sh.
2. Run the platform – After containers are up, open http://localhost:8001.
3. Develop locally –
   • Backend: mvn clean install (full or incremental) then bash docker/dev-app-builder.sh to hot‑reload.
   • Frontend: cd frontend && npm install && npm run build:prod (or npm run start for dev mode). The agent‑flow sub‑project builds the ELSA‑based flow editor.
4. Configure a model – Via the UI’s Model Configuration app, add the LLM endpoint and API key.
5. Create an app – Use the App Development page to start a blank app, select the configured model, design the workflow with ELSA, and test the chat interface.

Security note – The open‑source deployment does not include authentication for plugin uploads; exposing the service publicly may allow malicious plugins. Users are expected to add their own auth layer.

Who might use it

• Product managers who want to prototype AI features quickly.
• Engineers building internal AI tools that need a repeatable, version‑controlled workflow.
• Teams that want a unified store for LLM‑driven functions, RAG pipelines, or autonomous agents.

Where to find more – Full docs live in the docs/ folder, contribution guidelines in CONTRIBUTING.md, and the underlying FIT/Waterflow/ELSA frameworks are linked throughout the README.