AndersonBY/vector-vein

VectorVein – No‑code AI workflow builder

What it is – VectorVein is a desktop application that lets you design and run AI‑powered automation workflows without writing code. It combines large‑language‑model (LLM) calls, embeddings, speech‑to‑text, and other AI services into a drag‑and‑drop visual editor, similar to LangChain/LangFlow but packaged as a ready‑to‑run program.

Key capabilities

• Visual workflow editor – nodes representing LLM calls, embeddings, file I/O, web crawling, etc. can be placed on a canvas and wired together.
• Multiple LLM back‑ends – configure any OpenAI‑compatible endpoint (OpenAI, Azure, Ollama, LM‑Studio, vLLM, etc.). Separate endpoint and model settings let you mix providers.
• Embedding support – use OpenAI embeddings or connect a local service such as HuggingFace text‑embeddings‑inference.
• Speech recognition – plug an OpenAI‑compatible ASR service (e.g., Groq) and launch voice‑activated shortcuts that capture a screenshot and send it to the agent.
• Local FastAPI API – a built‑in REST server (http://localhost:8787) exposes endpoints to list, inspect, and run workflows programmatically.
• Templates & tutorials – official templates and an interactive online tutorial help new users build common flows (e.g., web‑article summarisation → mind‑map).
• Cross‑platform desktop – packaged with pywebview (WebView2 on Windows) and distributed as a single executable for Windows, macOS, and Linux.

How to get started

1. Download the latest release from the GitHub Releases page. The zip extracts a data folder for the SQLite database and static assets.
2. Install WebView2 runtime (Windows) if the app does not start.
3. Configure LLM endpoints in the Settings → API Endpoint tab. Add your OpenAI‑compatible URL (e.g., http://localhost:11434/v1/ for Ollama) and then select or add models under Remote LLMs or Custom LLMs.
4. Add an embedding backend if you need vector search, using the same endpoint‑configuration UI.
5. Create a workflow – drag nodes from the left pane onto the canvas, connect them, set node parameters, and save.
6. Run the workflow from the user‑interface tab or call it via the local API:

   import requests
   resp = requests.post('http://localhost:8787/api/workflow/run', json={
       'wid': 'my-workflow-id',
       'input_fields': [{'node_id':'node1','field_name':'input','value':'Hello'}],
       'wait_for_completion': True
   })
   print(resp.json()['data'])
   

Development notes

• Backend: Python 3.8‑3.11, managed with PDM. Run pdm install then pdm run dev to start the FastAPI server.
• Frontend: Vue 3 + Vite. Install with pnpm install and build with pnpm exec vite build (or use the provided shortcut pdm run build-front).
• Packaging: pdm run build creates a standalone executable via pyinstaller.
• Database migrations are handled by a small helper in backend/models; after schema changes run the migration script as described.
• CI/CD: GitHub Actions automatically builds Windows/macOS/Linux ZIPs for tagged releases.

Who it’s for – Users who want to harness LLMs, embeddings, or speech‑to‑text in repeatable automations but lack programming skills, as well as developers who need a quick visual prototyping layer before embedding the logic into their own code.

License – Open‑source, free for personal non‑commercial use (see LICENSE.md).