Describe a task. Romi builds the dataset — diverse, structured, ready to fine-tune.
One prompt in, a complete training set out. No scraping, no manual labeling, no busywork.
Up to 50,000 diverse, de-duplicated examples per dataset, streamed live as Romi generates them.
Backed by a state-of-the-art model with a vast context window for nuanced, high-quality examples.
Export to JSONL, Alpaca, ShareGPT, OpenAI messages, or CSV — one click, no reformatting.
Tell Romi the task in plain English — a domain, a tone, a label scheme.
Romi streams varied, high-quality examples in real time.
Pick a format and download, ready for your fine-tuning pipeline.
Every great model starts with great data — and gathering it is the part nobody enjoys. Apex exists to remove that friction. Romi is our first model: a focused engine that generates the training data your projects depend on, fast and on your terms. Spin up a dataset in the generator, compare plans and limits, or see how it works.
Romi outputs the formats teams already fine-tune with, like Stanford Alpaca and the Hugging Face training stack.