Clientforce Ai
Products▼ Solutions▼ Customers Pricing
Request access
←Back to menu
{{ col.h }}
←Back to menu
⛨ GUARDRAILS AND SECURITY

An agent you cannot talk out of the rules.

Most AI tools put their safety rules in the prompt. That means the rules are a suggestion — a persuasive customer, an odd phrasing, a long conversation, and the model talks itself into a discount you never authorised.

Ours live outside the model, at the gate every message passes through. She cannot argue with code she never sees.

HELD TODAY 37 of 1,284 checked
WHY THIS IS DIFFERENT

A rule in a prompt is a wish. A rule in the pipeline is a rule.

This is the single most important architectural decision in the product, and it is invisible in a demo. Here is what it means in practice.

✕ THE COMMON APPROACH
Rules inside the prompt
The instructions say “never offer a discount.” The model reads that alongside three thousand other words and a persuasive customer, and weighs it like everything else.
✕
A long conversation drifts
By message nine the early instruction carries less weight than the last thing the customer said.
✕
Phrasing can get around it
“What would you do for a returning customer?” is not the word discount, and the model helps.
✕
You find out afterwards
Usually from the customer, quoting a price you never approved.
✕
Every model update is a risk
Behaviour that held last month may not hold after the next release.
⛨ HOW WE BUILT IT
Rules at the send boundary
She drafts freely. Then every message passes through code that checks it against your rules before it can leave. The check is not a conversation and cannot be negotiated with.
✓
Length does not matter
The check on message nine is byte-for-byte the check on message one.
✓
Phrasing does not matter
It is not reading intent. It is matching your stated restrictions against the draft.
✓
You find out before, not after
A blocked message becomes a card in your needs-you list with the rule that stopped it.
✓
Model updates are contained
Swap the model underneath and the boundary behaves identically.
!
The test to run on any AI tool you are evaluating
Tell it never to discount. Then spend ten messages as a difficult customer asking for one, in every phrasing you can think of. If it eventually bends, its safety rules are in the prompt — and the same thing will happen to a real customer when nobody is watching.
WHAT YOU CONTROL

Six restrictions, set once, obeyed everywhere.

Set them at the workspace and every campaign inherits them. A campaign can be stricter than the workspace; it can never be looser. That direction is the whole design.

◷
Quiet hours, in their timezone
Nothing sends outside the window you set — calculated from where the contact is, not where you are. A customer two states over does not get a 6am text because it was 9am for you.
set once, inherited everywhere
▤
Daily and frequency caps
A ceiling per domain per day, and a separate limit on how often one person can be contacted across every campaign at once. Three campaigns that all include someone still only reach them once.
both enforced together
✕
Claims you will never make
Guaranteed outcomes, medical or legal promises, competitor comparisons, anything your regulator would object to. She cannot write them even when a customer asks directly.
blocked at the boundary
$
Prices and discounts
She quotes only what exists in your Business Core. Discounting is a switch, off by default, and no conversation can turn it on.
no invented numbers
◎
Tone and vocabulary
The words that are yours and the ones that are not. Formality, warmth, and the phrases you never want associated with the business.
matched, then checked
✓
Approval before first send
On by default for new campaigns. Nothing reaches a customer on a plan you have not read, and you can require it permanently for high-value contacts.
human in the loop
COMPLIANCE AND DATA

The obligations are yours. The plumbing is ours.

GDPR, CCPA, TCPA and CAN-SPAM are not features you switch on. They are constraints on how the system is built — where consent lives, how deletion propagates, and what an auditor can see without our help.

We are not your lawyers and this page is not legal advice. What we can do is make the compliant path the default one.

100% of sends carry a consent record and an audit entry
CONSENT ON RECORD
One opt-out stops every channel at once.
SECURITY

You are handing an agent your customer list. Ask hard questions.

These are the answers we would want before trusting anyone with ours.

⛁
Encrypted in transit and at rest
TLS 1.3 on every connection, AES-256 on stored data. Credentials for connected tools are encrypted separately with per-workspace keys, so a leak of one is not a leak of all.
◉
Roles that actually restrict
Owner, admin, agent and viewer. A viewer cannot send anything or spend a credit. An agent works inside your guardrails and caps and cannot raise them. Enforced server-side, not hidden in the UI.
⇄
Least-privilege integrations
Read-only where reading is enough. The ad platforms get analysis access, not spend access. Every scope is listed before you approve it and revocable in one click.
✕
Never used to train
Your contacts, conversations and outcomes are not training data for anyone — not us, not a model vendor, not another customer. It is contractual, not a policy page.
▤
Exportable on demand
Contacts, conversations, outcomes and the full ledger, as CSV, whenever you want. No support ticket, no retention negotiation.
◫
Durable execution
Campaigns run on a workflow engine that survives restarts, so a deploy mid-campaign cannot double-send or silently drop a step.
◷
On certifications, honestly
We are early, and SOC 2 Type II is in progress rather than complete. We would rather tell you that than imply a badge we do not hold yet. If your procurement needs the report before you can sign, talk to us and we will tell you where we are and when.

Set the rules once. Stop worrying about them.

Quiet hours, caps, claims, prices and approvals — configured in ten minutes and enforced on every message after that, on every channel.

Request access How we protect your domain
Clientforce Ai
An AI revenue workforce you hire against an outcome.
All systems operational
COMPANION APP
{{ f.h }}
© 2026 Clientforce Ai Privacy Terms Status