Guardrails

Guardrails let you define defensive behavior rules for the agent: which topics or responses it must avoid, how many violations it tolerates per conversation and per customer, and what message to show when a rule is broken.

The workspace-level whitelist of allowed terms is managed separately — see Guardrails whitelist .

Get Guardrails

GET https://api.platica.mx/v1/agents/{agentId}/guardrails

URL parameters

Parameter	Type	Description	Required
`agentId`	string	Unique identifier of the agent	✓

Response

{
  "enabled": true,
  "rules": [
    "No prometas reembolsos automáticos",
    "No menciones precios fuera de la lista oficial"
  ],
  "message": "Lo siento, no puedo ayudarte con eso.",
  "conversationStrikeThreshold": 3,
  "clientStrikeThreshold": 5
}

Response fields

Field	Description
`enabled`	Whether guardrails are active
`rules`	List of natural-language rules (maximum 20)
`message`	Message sent to the customer when a rule is violated
`conversationStrikeThreshold`	Maximum violations allowed in a single conversation (1-10)
`clientStrikeThreshold`	Maximum violations allowed historically per customer (1-10)

Update Guardrails

Every field is optional — only the ones present in the body are updated.

PATCH https://api.platica.mx/v1/agents/{agentId}/guardrails

URL parameters

Parameter	Type	Description	Required
`agentId`	string	Unique identifier of the agent	✓

Request body

{
  "enabled": true,
  "rules": [
    "No prometas reembolsos automáticos",
    "No menciones precios fuera de la lista oficial"
  ],
  "message": "Lo siento, no puedo ayudarte con eso.",
  "conversationStrikeThreshold": 3,
  "clientStrikeThreshold": 5
}

Parameter	Type	Description	Validation
`enabled`	boolean	Turn guardrails on/off	—
`rules`	array of strings	Natural-language rules	Maximum 20 entries, none empty
`message`	string	Message sent to the customer when a rule is violated	Maximum 500 characters
`conversationStrikeThreshold`	integer	Maximum violations per conversation	Integer between 1 and 10
`clientStrikeThreshold`	integer	Maximum violations per customer	Integer between 1 and 10

Response

{
  "status": "success",
  "message": "Agent guardrails updated successfully",
  "data": {
    "enabled": true,
    "rules": [
      "No prometas reembolsos automáticos",
      "No menciones precios fuera de la lista oficial"
    ],
    "message": "Lo siento, no puedo ayudarte con eso.",
    "conversationStrikeThreshold": 3,
    "clientStrikeThreshold": 5
  }
}

data always reflects the full guardrails state after the merge, not just the updated fields.

Previous Sub-agents Next Assist Rules