Skip to main content

​
When to use

  • Your app already calls the LLM and gets an answer
  • You want independent governance without rearchitecting
  • You need the lowest latency (~150ms p95)

​
Example

# 1. Your existing LLM call
answer = my_llm.chat("What is our MRR?")
tool_data = our_billing_api.get_mrr()

# 2. Add PromptWall verification
from promptwall import PromptWall
pw = PromptWall(api_key="pk_...")

result = pw.verify(
    prompt=user_question,
    answer=answer,
    tool_result=tool_data,
)

if result.governance == "block":
    return "Sorry, I can't verify that information."

return result.answer  # may be rewritten with caveats

​
What Verify checks

  • Evidence consistency — does the answer match tool_result?
  • Security — canary words, secret patterns, PII leaks
  • Mismatch type — contradiction / numeric / insufficient evidence
  • Confidence — high / medium / low

​
What Verify does NOT do

  • Does NOT call an LLM (no latency, no token cost)
  • Does NOT ground answers against external sources (you supply them)
  • Does NOT rewrite unless Security policy triggers
For rewrites / regeneration, use /v1/chat instead.