Generating high-quality synthetic data typically means staring at YAML files and praying your indentation is correct. ServiceNow just made that pain optional. They’ve open-sourced SyGra Studio, a visual workspace bolted onto their existing SyGra framework that lets you build, debug, and execute data-generation pipelines through a drag-and-drop canvas — no manual wiring required.
Think of it as a Node-RED for LLM workflows. You start by connecting a data source. Studio currently supports Hugging Face, local file systems, and ServiceNow instances. Once you point it at a dataset and preview a few rows, column names instantly become state variables you can drop into any prompt by typing an open curly brace. It’s a small detail that eliminates the guesswork of remembering variable names across a sprawling pipeline.
From there, you’re building flows visually. Drag an LLM node onto the canvas, pick a model — OpenAI, Azure, Ollama, Bedrock, Vertex, and vLLM are all supported — write your prompt, and define the output. Want a story generator that also summarizes itself? Two nodes, one feeding the other through a shared story_body variable. Toggle structured outputs if you need clean JSON, attach tools for function calling, or drop in Lambda and Subgraph nodes when things get complex. A built-in code panel shows the YAML or JSON artifact being generated in real time, so what you see is exactly what gets version-controlled.
Execution is where this gets genuinely useful for teams. Hit run and watch node-level progress stream in, with per-run tracking for token cost, latency, and guardrail outcomes. The system writes execution history to .executions/, giving you an audit trail and the ability to compare outputs across runs. ServiceNow includes a pre-built workflow that ingests the Glaive code-assistant dataset, generates answers, critiques them in a loop until the reviewer is satisfied, and produces fine-tuning-ready data. The whole thing is a git clone and a single make studio command away.
