ControlRules kept in code
The AI does only what you allowedRules, knowledge and approvals for an AI agent
It answers from what you gave it, follows your rules however hard a customer pushes, and asks a person before anything that costs money.
- Rules in plain words
- Approvals for money
- Every reply explainable
Three checks
Checked outside the AI, every time
A rule in a prompt is a suggestion. These are checks in code the AI can't talk its way past.
Your rules, first
Each step it wants to take is checked against your rules before it runs.
A person says yes
A campaign, a bigger budget or a price change waits as a request with the number attached.
The reply is checked
A reply that fails is held back, kept for you to read, and the chat comes to your team.
Approve mode
Read every reply before it goes
New number, or a business where one wrong word matters? The AI drafts, a person approves or edits, then it sends. Switch it per chat.
Explainable
Open any reply and see why
What it read, which rule applied and what it decided, for every reply in the last two weeks. A surprising answer takes a minute to fix.
And the rest
Small rules that add up to trust
AI guardrails, in detail
Honest about being an AI
Knowledge you can see
Test it first
Automations in plain steps
Read next
An AI you stay in charge of
Start freeFree plan, no card. WhatsApp at Meta's own rate.