FlowDrop Workflow Specification 1.0
The FlowDrop workflow specification is now 1.0: 386 rules for how a workflow is written, stored, validated and run, with 41 rulings and 53 conformance fixtures. Here's what 1.0 means and how to use it.
From the FlowDrop team
Updates, tutorials, and deep-dives on building AI workflow tooling.
RSS FeedThe Console: a command line for your canvas
Press one key and the FlowDrop editor gives you a command line. Here's what it does, why one typed line beats hundreds of lines of JSON and a dozen mouse moves, and why it helps AI build workflows that actually work.
Watch your workflow think
A first look at the redesigned FlowDrop editor. Test right where you build, watch each run light up node by node, and fix what broke without leaving the canvas. It's planned for the next minor release, and we'd love your feedback.
One name, several pieces: what FlowDrop is and who each part is for
FlowDrop started as a Drupal UI, became an npm editor, grew a Drupal engine, and gained a Rust backend. Here's how it got here, and which part you actually need.
100% test coverage, and still wrong
We've published the FlowDrop Workflow Specification: 399 rules for how a workflow is written, stored, validated and run, independent of any one implementation. Here's why we wrote it down, and what it changes for you.
The hidden cost of switching: when your agent's edge lives in someone else's database
The thing that makes your agent good isn't the model — it's the memory, context, and learned behavior it accumulates over time. When all of that lives in a vendor's database, switching doesn't just move your data. It resets your agent to day one.
GDPR for AI agents: can you delete data you don't control?
A customer asks you to delete their data. Your retention policy says conversations expire after 90 days. A legal hold says keep some of them. None of that is your call if your agent's data lives in a vendor's black box. Here's the fix.
The EU AI Act says you need human oversight. A system prompt isn't it.
The EU AI Act asks for human oversight and record-keeping over high-risk AI systems. If your only control is an instruction in a prompt, you don't have an answer. Here's what real oversight looks like in an agent.
Your auditor is going to ask where the agent's decisions are logged. Do you have an answer?
Before an AI agent goes anywhere near production, someone will ask for the record of what it decided and did. If that record lives in a vendor's system, you get whatever they choose to give you. Here's why the audit trail has to be yours.
Every conversation your agent has is your data. Do you own it?
An AI agent's memory is often the most sensitive data it touches — and the part teams control least. Here's the problem with renting your agent's memory, and what owning it changes.
Guardrails the model can't talk its way around
Prompts are suggestions. When you need an AI agent to actually stay within limits, the rules have to live outside the model — in the system that runs its actions. Here's how that works in FlowDrop.
When an AI agent calls its own tools, you lose control of what it does
The moment you let an AI model run its own tools, the decision and the action happen in the same breath — with no chance to check it. Here's the problem that creates, and how FlowDrop closes the gap.
Introducing FlowDrop 1.0
A visual workflow editor that your backend can own. Today we're releasing FlowDrop 1.0, a drop-in, backend-agnostic component for building AI workflow UIs.
Why your AI workflow tool should be backend-agnostic
Most AI workflow tools ship with their own cloud backend. Here's why that's a problem, and what a better approach looks like.