opensourceprojects.dev

A broadsheet for software that doesn't ask for your email

A workflow framework with a canvas, node forms, and variable scope built in
GitHub RepoViews0

Project Description

View on GitHub

Building an AI Workflow Platform? FlowGram Gives You the Canvas, Forms, and Variable Scope Already Done

You've probably had this thought: "I could build a visual workflow builder for our AI product." Then you start listing what it actually takes—a draggable canvas, node configuration forms, a variable scope system that resolves references across nodes, plus the actual node types like LLM calls and conditionals. That's months of infrastructure before you ship anything users care about. FlowGram is a framework from ByteDance that hands you those building blocks so you can focus on what makes your workflow platform different.

What It Does

FlowGram is a composable, visual workflow development framework and toolkit. It's aimed squarely at developers building AI workflow platforms, and it ships with the pieces you'd otherwise build from scratch: a flow canvas, a node configuration form, a variable scope chain, and ready-to-use materials like LLM, Condition, and Code Editor nodes. The framework packages are the building blocks you compose into your own platform.

There's also FlowGram Desktop, a native Agent workspace for macOS and Windows that brings those pieces together. You describe a task in chat and refine it on the canvas—adjusting nodes, conditions, parallel branches, joins, and loops. You can test individual nodes, inspect inputs and outputs, and revisit execution paths and historical workflow snapshots. Workflows can be published with versions for HTTP and MCP consumers, or called from other workflows. It's organized around project-based workflows and chats, with import/export, English and Simplified Chinese support, light and dark themes, and six industry example projects to start from.

Why It's Cool

  • It solves the boring 80%. The canvas, the forms, the variable scope chain—these are exactly the parts that are tedious to build and hard to get right. Variable scope in particular is the kind of thing that seems simple until you're three levels of nesting deep and references stop resolving. Having it built in is a real head start.

  • The toolkit and the app are separate things. The framework packages are building blocks for your own platform; FlowGram Desktop is a reference implementation that shows them working together. That separation matters—you're not adopting an opinionated product, you're adopting a set of primitives you can assemble your own way.

  • The desktop app doubles as a debugging environment. Testing individual nodes, inspecting inputs and outputs, and stepping back through execution paths and historical snapshots is the kind of tooling that usually comes late, if at all. Having it in the preview means you can see how workflows actually behave, not just how they look.

  • Publishing for HTTP and MCP consumers is a smart bet. Versioned workflows that other systems (or other workflows) can call turns your builder from a diagram tool into actual infrastructure. MCP support in particular signals where the ecosystem is heading.

  • It's honest about being a preview. The installers are unsigned and the macOS builds aren't notarized. That's not a selling point, but it tells you where the project is in its lifecycle—and the install script handles the quarantine flag so you're not fighting Gatekeeper.

How to Try It

The fastest path is the desktop preview. On macOS (Apple Silicon and Intel), paste this into Terminal:

curl --fail --silent --show-error https://flowgram.ai/desktop/install.sh | sh

The script downloads the latest preview, verifies its SHA-256 checksum, and installs FlowGram. It removes quarantine only from the installed app so the unsigned preview can open, and it doesn't need sudo. Run the same command to update (quit FlowGram first).

If you'd rather grab an installer directly, there are permanent links for each platform—no GitHub sign-in required:

  1. macOS · Apple Silicon (ARM64): download the DMG from the release page.
  2. macOS · Intel: download the DMG from the release page.
  3. Windows 10/11 · x64: download and run the setup EXE.

Checksums and build information live on the release page. Open the .dmg on macOS or run the .exe on Windows.

If you're evaluating the framework itself, start at flowgram.ai and the repository, where you'll also find the demo sandboxes (CodeSandbox and StackBlitz) linked in the README.

Final Thoughts

FlowGram is best suited to teams building an AI workflow platform who'd rather not spend the first few months on canvas mechanics and scope resolution. The framework's built-in materials and the desktop app's debugging and publishing story make it a credible starting point, though the preview status—unsigned installers, notarization still pending—means you should treat the desktop app as something to explore rather than depend on. If the scope chain and node forms are the parts of your roadmap you dread most, this is worth an afternoon.


Follow @githubprojects for more developer tools and open source projects.

Back to Projects
Project ID: flowgram-aiLast updated: November 19, 2026 at 07:04 AM