Dashboard — WonderwallAi

Requests by Endpoint

Topic Routing

Allowed Topics

Define allowed conversation topics. Messages outside these topics will be blocked by the Semantic Router.

Similarity Threshold

0.35

Lower = stricter matching. Default: 0.35

Sentinel LLM Classifier

Sentinel Enabled

LLM-based injection detection for ambiguous messages

Sentinel Model

General Settings

Bot Description

Canary Prefix

Fail Open

Allow messages through when the firewall encounters an error

Block Messages

Off-Topic Block Message

Injection Block Message

Your Plan

Available Plans

Generate Canary Token

Canary tokens detect prompt extraction attacks. Inject the token into your system prompt, and WonderwallAi's egress filter will block any response that leaks it.

Session ID

Canary Token

Prompt Block (add to your system prompt)

Python SDK

Install

pip install wonderwallai

Scan Inbound Messages

from wonderwallai import WonderwallClient

client = WonderwallClient(
    api_key="your_api_key",
    topics=["customer support", "product questions"],
)

# Scan before sending to LLM
verdict = client.scan_inbound("How do I return my order?")

if verdict.allowed:
    # Safe to send to LLM
    response = call_your_llm(message)
else:
    # Message was blocked
    print(verdict.message)

Scan Outbound Responses

# Check LLM output for leaked data
verdict = client.scan_outbound(
    text=llm_response,
    canary_token="WONDERWALL-abc123"
)

if verdict.action == "redact":
    # PII was found and redacted
    safe_response = verdict.message
elif not verdict.allowed:
    # Canary token leaked — block response
    safe_response = "I can't share that information."

Hosted API (cURL)

Scan Inbound

curl -X POST https://wonderwallai-production.up.railway.app/v1/scan/inbound \
  -H "Authorization: Bearer your_api_key" \
  -H "Content-Type: application/json" \
  -d '{"message": "How do I return my order?"}'