Toolkits bundle scanners for secrets, PII, injection, toxicity and schema validation that you can chain in front of or behind a model. Pick ones you can self host and extend with your own validators.
Read in this order
- Open sourceLLM Guard (Protect AI)Start here: a ready set of input and output scanners to chain.
- Open sourceGuardrails AIValidator-based checks with structured output enforcement.
- Open sourceOpenAI Guardrails for PythonA drop-in option if you already use OpenAI clients.