DEV Community

Cover image for I built an open-source visual AI agent builder — PandaFlow
Ajay Kumar
Ajay Kumar

Posted on

I built an open-source visual AI agent builder — PandaFlow

#ai #agents

If you've ever tried to wire up an AI agent — connecting an LLM to a database, a webhook, some Python code, and an API — you know how fast it becomes spaghetti.

That's why I built PandaFlow (https://github.com/pandastack-io/pandaflow) — an open-source, drag-and-drop visual workflow builder for AI agents.

What it does

You connect nodes on a canvas. Each node can be:

  • An LLM call (GPT, Claude, etc.)
  • Python / Node.js / Go / Bash code — running in isolated microVMs
  • A database query, API call, or webhook trigger
  • A transformation, condition, or template

160+ nodes out of the box. Real-time execution monitoring via SSE. Encrypted secrets per org.

The stack

Next.js 16 · React 19 · TypeScript · Drizzle ORM · PostgreSQL · Redis · React Flow · NextAuth v5

Code runs in Sandflare.io (https://sandflare.io) microVMs — fully isolated, no shared runtimes.

Run it locally in 5 commands

git clone https://github.com/pandastack-io/pandaflow.git
cd pandaflow
npm install
cp .env.example .env.local   # set SANDFLARE_API_KEY=mock-api-key for local dev
npm run db:migrate && npm run dev
Enter fullscreen mode Exit fullscreen mode

Open http://localhost:3000 — done. No cloud account needed.

Why open source?

Because AI agent tooling shouldn't be locked behind a SaaS paywall. PandaFlow is MIT licensed — self-host it, fork it, extend it.

⭐ If this is useful to you, a star on GitHub goes a long way: github.com/pandastack-io/pandaflow (https://github.com/pandastack-io/pandaflow)

Contributions, issues, and ideas very welcome!

Top comments (0)