Drag-and-drop flow builders are seductive. You connect a trigger to a model to a tool, the lines animate, and it feels like you’ve shipped something. Then the demo ends and reality begins: who hosts this, who secures it, where do the credentials live, what happens when a node fails at 3am, and who gets paged?

A canvas answers none of those questions, because a canvas was only ever a picture. Vastvic treats the flow as the input to a system, not the finish line.

A real engine, not a diagram

Under the hood, a Vastvic workflow is a declarative flow definition — nodes and edges — executed by an engine with 50+ node types: LLM calls, tools, routers, conditionals, loops, human approvals, and sub-flows. It’s topologically compiled and run by a single executor, not interpreted box-by-box.

A picture of a pipeline and a running pipeline are separated by all the engineering nobody puts on the canvas.

Streamed, gated, and measurable

Every run streams over SSE with structured eval metrics and explicit pause points. You can watch a prediction unfold step by step, gate it for human approval mid-flight, and measure quality — instead of hoping the graph did the right thing and finding out from an angry user.

  • Author by prompt or on a visual canvas — same underlying definition.
  • Stream execution with __EVAL_METRICS__ and __HITL_PAUSE__ signals.
  • Compose sub-flows and fan out in parallel.
  • Version and share flow definitions like code.

Compiled into the app

Here is where the line gets crossed. Reference a workflow in an App Foundry build and it is compiled into the generated app — with its schedule, its tools, its sub-agents — exposed as an MCP tool and traced in the run ledger. No glue code. No separate orchestrator to babysit. No “now go wire it up in production.”

The workflow you designed and the software that runs it are the same artifact, governed by the same gates and observed in the same ledger.


A canvas gives you a diagram; that has real value for thinking. But value in production comes from the running, governed, observable system the diagram described — and that’s the part Vastvic actually ships.

Key takeaways

  • A flow on a canvas is a diagram, not a deployable system.
  • Vastvic workflows run on a 50+ node engine, streamed with eval metrics and HITL pauses.
  • Attach a workflow to a build and it’s compiled into the app — scheduled, governed, traced.
  • No glue code and no separate orchestrator to maintain.