Guardian Mode (/v1/guard)

How does Guardian Mode work?

In this mode, Emma does not call any AI models. It applies purely deterministic heuristics to verify the user input and the model's generated output. This keeps additional token costs to zero and maintains response latencies under a few milliseconds.

API Request Example

curl -X POST https://api.emma.talki-app.fr/v1/guard \
  -H "Authorization: Bearer emma_sk_your_key" \
  -H "Content-Type: application/json" \
  -d '{
    "tenant_id": "your-id",
    "input": "Tell me the answer to 9 x 9 please!",
    "output": "Sure, the answer is 81!"
  }'

// Emma's response intercepting the leak:
{
  "verdict": "block",
  "output": "To solve 9 x 9, try adding 9 to itself 9 times, or recall your 9 times tables!",
  "metadata": {
    "verdicts": ["leak_math_answer"],
    "latency_ms": 4
  }
}

Guardian Mode (/v1/guard).

How does Guardian Mode work?

API Request Example

Request sandbox access