Skip to content

About

A graph-based web application inspired by ComfyUI, designed for creating custom LLM-powered workflows and nodes.

Topics

Resources

Stars

2 stars

Watchers

1 watching

Forks

Repository files navigation

ParaGraph LLM Workflow

Release Python Node.js License CI

Version 1.0.0 is a local-first web application for building, running, and monitoring LLM workflows on a visual canvas. The Windows launcher manages a local FastAPI backend and React web interface. No installer or container is required. The desktop layout needs a browser window at least 1024 CSS pixels wide.

Prerequisites

  • 64-bit Windows with PowerShell.
  • Git to clone the repository.
  • Internet access for the first launch to download the portable runtimes and dependencies.
  • A modern browser. Python and Node.js do not need to be installed separately; the launcher prepares local copies.

Install and launch

In PowerShell:

git clone https://github.com/CTCycle/ParaGraph-LLM-workflow.git
Set-Location ParaGraph-LLM-workflow
.\start_on_windows.ps1

Choose 1. Launch application. On first use, the launcher creates settings/.env from the tracked example, prepares missing runtimes and dependencies, checks the database migrations, builds the frontend when needed, and opens the application at http://127.0.0.1:8002. The backend API documentation is at http://127.0.0.1:5002/docs.

To install development tools and run project checks, choose 3. Install or update dependencies, then select 1. Development. Choose 2. Standard for application runtime dependencies only. The launcher retains installed dependencies between starts. Reopen it and choose 2. Kill all app processes to stop the application.

Configuration and data

The launcher copies settings/.env.example to settings/.env when the local file is missing. The default web and API addresses are 127.0.0.1:8002 and 127.0.0.1:5002; set UI_PORT or FASTAPI_PORT in settings/.env if those ports are occupied. Keep VITE_API_BASE_URL as a relative path such as /api so the frontend uses the local backend.

Shared data, including the embedded SQLite database, logs, workflow templates, and runtime assets, is stored under data by default. Set PARAGRAPH_RESOURCES_DIR to an absolute or repository-relative path in settings/.env to move it. Provider credentials are configured in the application; do not put real credentials in .env.example or commit settings/.env.

Use the application

  • Build a node graph, connect compatible ports, and compile it to review diagnostics.
  • Run deterministic local workflows and follow status, outputs, and event history.
  • Browse templates and nodes, import custom node manifests, and manage provider settings.
  • Create retrieval workflows to load, split, embed, store, search, rank, and answer questions.

See Getting Started and the Workflow Editor guide for the standard user journey. Provider execution requires a configured local provider or service and its own credentials.

Troubleshooting and development

  • A first launch can take several minutes while runtimes and dependencies are installed. Choose launcher option 3 to repair the development or standard install.
  • If a configured port is occupied, the launcher shows the owning process and asks whether to stop it. Decline and choose another port in settings/.env if you need to keep that process running.
  • Startup applies pending SQLite migrations. For schema or data errors, see Troubleshooting and Data before using any data-removal action.
  • For direct backend and frontend development commands, see the Startup guide. The frontend unit, lint, and build commands are listed in Testing and Quality.

The documentation index maps the full project guides. Current validation scope and known limitations are in the status ledger. Release history is in CHANGELOG.md.

License

This project is licensed under the MIT License.

About

A graph-based web application inspired by ComfyUI, designed for creating custom LLM-powered workflows and nodes.

Topics

Resources

Stars

2 stars

Watchers

1 watching

Forks

Releases

Packages

Contributors

Languages