Forja
Hard rules in the brain
The bot is forbidden from stating prices, services, or hours that aren't in your business info. It's not a tone suggestion — it's a system rule.
🛡️ Superpower · Anti-hallucination Shield
What if the bot makes up a price for a customer?
It's fear #1 before handing your WhatsApp over to an AI. A made-up discount or a fake business hour costs you money and trust. The Anti-hallucination Shield exists for exactly that reason — so it never happens.
› included with Forja + · in your cloud, with your data
01 · Hard rules02 · Searches before stating03 · Verifies before sending04 · Gap radar
Passages checked
Top 5
from your knowledge base
Confidence threshold
0.7
below that, it won't state it
Gap radar
30 days
of unanswered questions
// that's literally written in the bot's code — not fine print
How it works
01
02
03
04
Four locks, in series.
It's not a fancy prompt: it's hard rules, threshold-gated search, pre-send verification, and a radar that collects what was missing.
Searches before it states anything
Before giving out any fact, it checks your knowledge base: it pulls the 5 most relevant passages and only trusts them if they clear the 0.7 threshold. Below that, it won't state it.
Pre-send verifier
Every reply that states a fact — price, hours, a promise, a policy — is checked against your business info before it goes out. Without backup, the bot answers "let me confirm that" and hands you the conversation.
Knowledge Radar
Collects every question the bot couldn't answer over the last 30 days. You fill the gaps in a minute — and that question never comes up unanswered again.
The example
What it doesn't know, it won't invent.
The customer asks something that isn't in your business info. The bot doesn't guess — it admits it, alerts you, and the Radar logs it so you only have to fill the gap once.
// on Monday, the Radar shows you that 4 people asked the same thing — you add it once and it never has to guess again.