Skip to main content
The composer opens with a guided prompt: a single sentence that describes the run you are about to start.
The guided prompt
Each bracketed part is a slot you can click. The sentence is the only place a run is described, so what you read in the composer is exactly what the agent receives.

​
Filling it from the menus

Click any slot to open the run configuration panel at that slot. Pick a value and the panel moves to the next decision in the sentence, in reading order:
  1. Model: search Hugging Face or paste a repo id
  2. Serving engine, or leave it on Auto
  3. GPU, or leave it on Auto
  4. Output format, or leave it on Auto
After the last one the panel closes and your sentence is complete. You never reopen the menu to reach a decision you skipped. Only the model is required. Every other slot rides Auto, which means RunInfra picks the best option for your chosen setup. Auto is a real choice, not a deferral: the unfilled model slot is highlighted to show it is the one decision waiting on you, while the Auto slots stay quiet.
Filled slots render as tags with the same icons used everywhere else in the product, so a model, an engine, a GPU, and a format all read as one family.

​
Writing your own prompt instead

Select Remove template and the composer becomes a blank page. Describe the run however you like, in your own words. Free writing and the menus still work together:
  • Picking a model while you are typing inserts it as a tag at your cursor. Your text is never rewritten or replaced.
  • Picking anything from the menus while the composer is empty brings the guided sentence back with your pick already filled in.
The button next to the composer tells you which mode you are in. It reads Insert template when the composer is empty, and Use guided prompt when you have written something of your own, where it adds the guided sentence without touching your text.

​
Choosing where the run happens

One decision has no slot in the sentence: Runs on, which controls where the work is placed. It stays as a control below the composer, and the agent infers a sensible placement from your use case when you leave it alone.

​
What to do next

Prompting best practices

What to include so the agent builds the right pipeline the first time.

Example prompts

Complete prompts for common use cases.