Building blocks · by Meta
Prompt Guard
Catch hidden instructions in text before they reach your assistant.
text → data
- Way in
- Official APINeeds approval, other
- To start
- Freeopen weights (gated HF form)
- Runs
- On your MacAlso through groq, Hugging Face
- Checked
- Nov 4, 2026Prices as of that day
What it can do 1
Classify a text chunk as benign or a prompt attack (injection/jailbreak)512-token context: split long inputs and scan segmentstext → datafree locally; Groq $0.04/M tokens
Does the same job
Prices and limits as of Nov 4, 2026, from Meta’s own pages. They change; check before you rely on them.Not affiliated with Meta. Not advice.Something wrong?