Botvetter Guard · for enterprises · early access

Stop risky repliesbefore they send.

The free report tells you what your bot said last week. Guard is the enterprise tier: a realtime API that vets every reply before your customer sees it. Same six checks, moved from hindsight to the moment it matters.

Create your organization Start with the free report

ALLOW: Nothing risky found. Send the reply as-is — this is the verdict almost all replies get.
REVIEW: Something deserves a second look. Route the reply to a human, or send a safe fallback.
BLOCK: A made-up policy, an unapproved promise, a fake-human claim. Swap in an escalation instead.

Three steps to a guarded bot

Create your organization

Sign up in a minute — self-serve, no sales call. Create a workspace for your bot and the dashboard hands you a key with a ready-to-paste first request.

Check each reply before it sends

Your bot drafts a reply; your backend sends it to Botvetter with a little conversation context. One request, one verdict.

Act on the verdict

Allow means send it. Review means hold it for a human. Block means that reply would have been next week's incident — it never reaches the customer.

One request, one verdict

fast mode answers with the triage model in well under a second. deep mode adds a full audit with exact-quote evidence when triage escalates — worth it for review queues and high-stakes flows.

Request

curl https://judge.saivinod.in/api/guard/check \
  -H "Authorization: Bearer bv_YOUR_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "messages": [
      { "role": "customer", "text": "Can I return my kettle after 40 days?" }
    ],
    "reply": "I have extended your refund window to 90 days for you.",
    "mode": "fast"
  }'

Response

{
  "action": "block",
  "risk": "likely",
  "reasons": ["unauthorized_commitment"],
  "degraded": false,
  "latencyMs": 412
}

Set up in five minutes

Keys are stored as hashes — we cannot read them back, and neither can anyone who reads our database. Rotate or revoke them any time, per workspace.
Emails, phone numbers, card numbers, and addresses are redacted before any reply is analyzed or logged. Guard activity is deleted after 30 days.
Want the verdicts tuned to your real policies? Include your policy text in the request and the audit checks against it.

Your team's dashboard

One organization for your whole team: workspaces per bot, rotatable API keys, live API logs and analytics, role-based access, an audit trail, and alerting that pages you when the block rate spikes — and stands down by itself when it recovers.

Create your organization Sign in