Fusova
It asks before it answers.
Every time.
The only message that ever sends on its own is the friendly “we got your call” — it never quotes a price or books a job. Anything that actually answers your customer or commits your business waits for your one-tap approval. Edit, Discard, or Pause any draft. Prefer to approve even that first hello? Switch on Full-approval mode and nothing sends without you. Every Fusova tenant runs the same gates — banned-phrase blocking, PII redaction, confidence threshold, approval queue — before any substantive text reaches a real customer.
The guardrails
Seven gates between a draft and your customer.
A helper that texts your customers on your behalf has to be worth trusting. Here's exactly how that trust is built — in plain English.
Hard cost caps
Every tenant gets a monthly spending cap, set at the proxy layer. A runaway prompt loop, a hostile input, a workflow that misfires — none can blow past the ceiling. The cap is enforced before the call leaves the network. Not after the bill arrives.
Banned-phrase blocking (per tenant)
You define phrases Fusova is forbidden to use on your behalf. Promissory commitments, legal-risk language, pricing it shouldn't quote, competitor names. The check runs after the draft but before anything is sent. If a draft trips it, it routes to your approval queue instead of going out.
In-house transcription & PII redaction before the model sees it
Voicemails and missed-call transcripts run through our own in-house Whisper transcription, then a local model strips PII (social security numbers, full credit cards, addresses) before the prompt ever goes to the cloud model. Sensitive data never leaves your tenant for inference. The full unredacted transcript stays in your audit log.
Confidence threshold on every draft
Every draft comes with a self-rated confidence. A booking confirmation needs higher confidence than a "thanks for reaching out." Below the threshold, the draft routes to your approval queue instead of going autonomous. You set the threshold; the default is calibrated for missed-call recovery.
Approval-by-default for the first 7 days
Test Mode is on for every new pilot. Fusova processes real inbound but every outbound message waits for your one-tap approval. You see the work before it ships. Day 7 you flip to LIVE, or extend Test Mode another week. No rush.
Audit logs in plain English
Every decision Fusova makes is logged: what came in, what it considered, how confident it was, what passed or failed, what it sent. Plain English, not JSON. You can review any thread back to its trigger. Kept encrypted at rest.
SMS compliance, built in
Business texting is regulated — carriers require registered, consent-based messaging (A2P 10DLC). Fusova runs the text-back from a properly registered business number, honors STOP/opt-out automatically, and is built for the inbound, consent-based recovery side — never outbound cold-texting. That's how your number keeps its reputation with the carriers.
You can pause it with one text
Reply PAUSE to the Fusova number and everything stops in under a second. Reply RESUME when you're ready. Reply MINE to take over a single conversation while the rest keeps running. Owner authority is never more than one text away.
Why this is rare
Most AI tools don't ship a trust layer.
As of June 2026, the largest competitors in this space publish no mention of per-tenant banned-phrase blocking, local PII redaction, or approval-by-default on their security pages. The trust gap at the small-business tier is real.
| Guardrail | Fusova | AI assistant | AI receptionist |
|---|---|---|---|
| Substantive replies wait for owner approval | Yes — one-tap Send (only the “we got your call” ack auto-sends; Full-approval mode holds even that) | Often autonomous | Answers live, no approval |
| Banned-phrase blocking (per tenant) | Yes | No | No |
| Local PII redaction before model call | Yes | No | No |
| In-house Whisper transcription | Yes | Third-party API | Third-party API |
| Approval-by-default first 7 days | Yes — Test Mode | No | No |
| Plain-English audit log per thread | Yes | JSON traces in dashboard | Call recording only |
The competitor columns are shorthand for the category — checked against the largest published security pages in June 2026. Specific vendors evolve; we'll update when they catch up.
Two things we do differently
Compliant texting and our own transcription.
Two parts of the recovery workflow most tools outsource or hand-wave — carrier compliance and voicemail transcription — we run ourselves, on purpose.
US carriers regulate business texting through A2P 10DLC: a registered business identity, consent-based messaging, and automatic opt-out handling. Fusova runs the text-back from a properly registered number, honors STOP instantly, and never does outbound cold-texting. That's what keeps your number deliverable and your business in good standing with the carriers.
When a caller leaves a voicemail, the transcription runs on our own in-house Whisper — not a third-party transcription API your customer's words get shipped off to. It feeds the triage and stays in your audit log. One fewer outside service touching your callers' voices.
The bigger principle
Predictable AI. For unpredictable businesses.
Your dinner rush is unpredictable. Your peak season is unpredictable. The next emergency call is unpredictable. The one thing that shouldn't be unpredictable is the tool you trust to text your customers back. Fusova's job is to be the boring, reliable part of your operation. Not the unknown.
Read the audit log for yourself.
Start a pilot. We'll send you a sample log from the demo tenant before you sign up — redacted, real, plain English. No credit card to start.