Guardrails in SUPERWISE are validators that keep an agent's prompts and responses within a set of rules, so it produces appropriate, accurate, and safe content. Input guardrails check prompts before they are processed, and output guardrails check responses before they reach the user. A guardrail's rules, called guardrules, live in its versions. See Guardrail for the concept.
Endpoints
- List guardrails
- Create a guardrail
- Run guardrules
- Get the guardrule JSON schema
- Validate guardrules
- Get a guardrail
- Update a guardrail
- Delete a guardrail
- Search guardrails
- List guardrails with a tag
- Add tags to a guardrail
- Remove tags from a guardrail
- Run a bulk action on guardrails
- Archive a guardrail
- Unarchive a guardrail
- Create a dataset for a guardrail
- Verify a dataset schema for a guardrail