Nexxtool AI
I can give you 20% off your next order.
It offered a discount you don’t give.
Your rule: never say “% off”.
I can't offer a discount, but here's our current pricing page.
AI answers. We check.
Stop AI mistakes before they reach your customers.
Check every AI reply against your rules before it's sent. If it breaks one, your app gets a block result and the reason.
2,500 free checks a month for 3 months. Grand opening offer, sign up by February 2, 2027.
- Catch mistakes before sending
- Checked against your rules
- See why a reply was blocked
Works with any AI that writes text
- OpenAI
- Anthropic
- Google Gemini
- Meta Llama
- Mistral AI
- Your own model
What happens to every AI reply
Your AI writes a reply. Nexxtool AI checks it before your customer sees it.
Step 1: Your AI writes a reply
AII can give you 20% off your next order.
Step 2: Nexxtool AI checks it
- Banned wordspassed
- Linkspassed
- Knowledgefailed
Step 3: Problem found
Offered a discount you don’t give.
Step 4: Fixed automaticallyYour app gets the result + reason
Safe AI address onlyI can’t offer a discount, but here’s our current pricing page.POST /v1/validateStep 5: Your customer gets a safe replyYour app decides: send, retry or hand off
Only safe replies go out.Still failing after 2 tries? Your customer gets “Let me get a team member to help with that.”You choose what happens next.
What you get
Everything included. Tap any feature to see how it works.
Reply checking
Every AI reply is checked before your customer sees it. You get a clear pass or block straight away. Nothing goes out unchecked.
Your rules
Set the rules that matter to your business: banned words, required words, length, patterns and structure. Every reply is checked against them. Change a rule any time and the next check uses it.
Clear reasons
Every block names the rule that failed, so you know exactly what to fix. You can show the reason to your team or keep it in your logs. No guessing why a reply was stopped.
Try it live
Paste a reply and run a check in your browser. Add or change rules and see the result straight away. It is an easy way to see what Nexxtool AI does.
Block private data
A reply that contains a card number, a social security number or a secret key is stopped before your customer sees it, and the reason says what it found. With the Safe AI address, your AI is asked to take it out, and a safe fallback goes out if it still fails. It is on by default for new rules.
Auto-repair
Fixes small formatting mistakes for you instead of only blocking the reply. Your customer gets a working reply, not an error. A reply that still fails is blocked.
Safe AI address
Paste one address into your chatbot or tool. Every reply is checked, fixed if needed, and a safe fallback goes out if it still fails. No code needed.
Knowledge checks
Compare replies with your own policies and documents to catch wrong facts, such as a 90-day return window when your policy says 30. Each knowledge check also counts as 1 normal check.
Ready-made fact packs
Pre-built checks for common topics like refunds, shipping and pricing. Switch a pack on and replies are checked against it, so you do not have to write the facts yourself.
Bring your own AI
Use our checker, or connect your own AI key. You stay in control of which AI does the work.
Usage you control
Every plan has a hard monthly cap on checks. When you reach it you are not billed extra. See your usage at any time, so a busy month never comes as a surprise.
Check history
Every check is saved with its result and reason, so you can see what was caught and when. How far back you can look depends on your plan.
Keys you control
Create a key for each app or teammate, and revoke any key whenever you need to. You can have as many keys as you need.
Guided setup
Answer a few questions about your business and get your first rules, ready to use. It is the easiest way in, and you can change the rules afterwards.
Works with your chatbot
Paste your Safe AI address into any chatbot or tool that lets you set its AI address. No code needed. Replies are checked on the way through.
Try it live
Pick a mistake, edit the reply or the rules, then run the check. It uses the same rule format as the API and runs in your browser. No key needed.
textPress Run check to see the result
Raw JSON response
No developer? Change one address.
Paste your Safe AI address where your tool asks for its AI address. Every reply is checked, fixed if needed, and only safe replies go out.
AI address: https:/
AI address: https:/
Checked by Nexxtool AI
- Sign up
- Copy your Safe AI address
- Paste it into your tool
Who it's for
If AI writes to your customers or feeds another system, give it rules it can be checked against.
Online stores
Keep prices, links and wording on-script.
SaaS and support teams
Keep replies on-message, with fewer escalations and apologies.
Law firms
Check AI replies against your firm's own guidelines.
Clinics
Check AI replies against your approved wording.
Teams building on AI models
Add one check after your model call. Fail loudly and log everything.
Automation pipelines
Check structure, fields, palette, fonts and length of generated HTML, JSON and scripts before they reach downstream systems.
Nexxtool AI checks replies against the rules you set. It isn't legal or medical advice and doesn't make you compliant with any regulation.
Pick a plan by volume
Four plans, billed monthly, priced by how many replies you check.
Free
Try it out
$0/month
2,5001,000checks a month
$0 · 2,500 checks/month for your first 3 months, then 1,000/month
Start free- 1 connection, with auto-fixSafe AI address
- 50 a monthKnowledge checks on our AI
- 1Your own documents
- 7 daysHistory
- Email supportSupport
Starter
Small shops and side projects
$29/month
10,000checks a month
Choose Starter- 3 connections, with auto-fixSafe AI address
- 1,000 a monthKnowledge checks on our AI
- 10Your own documents
- 30 daysHistory
- Email supportSupport
Billed monthly. Renews automatically until you cancel. Cancel anytime in your dashboard.
Pro
Growing businesses
$99/month
100,000checks a month
Choose Pro- 10 connections, with auto-fixSafe AI address
- 10,000 a monthKnowledge checks on our AI
- 100Your own documents
- 90 daysHistory
- Priority email supportSupport
Billed monthly. Renews automatically until you cancel. Cancel anytime in your dashboard.
Scale
High volume
$299/month
1,000,000checks a month
Choose Scale- Unlimited connections, with auto-fixSafe AI address
- 35,000 a monthKnowledge checks on our AI
- 500Your own documents
- 1 yearHistory
- Priority email supportSupport
Billed monthly. Renews automatically until you cancel. Cancel anytime in your dashboard.
- Every plan includes
- Your own rules
- Unlimited API keys
- Auto-repair
- Guided setup
- All fact packs
- Hard monthly cap, no surprise bills
Requests per second per API key: Free 10 · Starter 50 · Pro 200 · Scale 1,000
Each knowledge check also counts as 1 normal check. With your own AI key, knowledge checks are limited only by your normal checks.
No credit card needed for the Free plan. Paid plans are billed monthly with a hard monthly cap, so there are no surprise bills. Cancel anytime from the dashboard. See full pricing
Your data
How we look after your account, in plain words.
Backups
We back up our database and test that we can restore it.
Hard caps
Your plan has a hard monthly limit. You are never charged extra for going over.
Keys you control
You can revoke any API key at any time. After that it stops working.
What we store
For each check we keep the result, the failed rules and the first 60 characters of the reply. The full reply isn’t stored.
Before you sign up
What counts as a check?
One AI reply checked once. Every call to the check endpoint counts as one check, whether the result is pass or block. Failed lookups (for example an unknown rule set) aren't billed, and a repair adds one more check.
Which AI does it work with?
Any AI that writes text. Nexxtool AI checks the reply text, not the model. OpenAI, Anthropic, Google, open-source: if your code has the reply as text, you can send it.
Do I need a developer?
No. The Safe AI address and guided setup let you start without writing code. Yes, if you want to use the API: a developer adds one call to check each reply.
Does it understand what a reply means?
Rule checks match the words and patterns you set, so a banned word catches that wording, not every way of saying it. Knowledge checks go further: they compare the reply to your own policies and docs to catch wrong facts.
What happens to a blocked reply?
With the API, your app gets the result and decides: send a safe fallback, retry, or hand off to a person. With the Safe AI address, Nexxtool AI can fix the reply automatically (up to 2 tries). If it still fails, your customer gets “Let me get a team member to help with that.”
Can I go over my limit?
No. Every plan has a hard monthly cap. Calls return 402 quota_exceeded until the 1st of next month (UTC) or until you upgrade. There are no silent overage charges. Each key also has a per-second rate limit: Free 10, Starter 50, Pro 200, Scale 1,000.
Can I cancel anytime?
Yes. In your dashboard: Plan & usage, then Manage subscription, which opens the customer portal where you can cancel. Cancellation takes effect at the end of the period you've paid for. Signup is open: create a free account on the sign-up page.
What are knowledge checks and fact packs?
Knowledge checks compare a reply with your own policies and documents to catch wrong facts, such as a refund window you don’t offer. Fact packs are ready-made sets of facts for common topics like refunds, shipping and pricing, so you don’t write them yourself.
What’s the grand opening offer?
2,500 checks a month for 3 months, for sign-ups from November 1, 2026 to February 2, 2027. After that, Free is 1,000 checks a month.
Is my data used to train AI?
No. We don’t use your data to train AI models, and we don’t sell it. Knowledge checks that use our AI run on Ollama Cloud. Ollama states that prompt and response data is never logged or used for training (source: ollama.com/cloud). If you connect your own AI key, your AI provider’s terms apply to that data. For each check, Nexxtool AI keeps the result, the rules that failed, the rules used, a fingerprint and the first 60 characters of the reply, for the history period of your plan. We don’t keep the full reply. Details are in our Privacy Policy at /privacy.
For your developers
Create a key in your dashboard, paste the curl, and your first check shows up in your history.
# Check an AI reply against a rule set
curl -X POST https://nexxtool.ai/v1/validate \
-H "Authorization: Bearer wtk_live_..." \
-H "Content-Type: application/json" \
-d '{
"output": "Order confirmed for John Doe.",
"contract": {
"type": "text",
"must_contain": ["Order confirmed"],
"min_length": 10
}
}'
# → { "pass": true, "score": 1, "issues": [] }const r = await fetch("https://nexxtool.ai/v1/validate", {
method: "POST",
headers: {
"Authorization": "Bearer " + process.env.NEXXTOOL_API_KEY,
"Content-Type": "application/json",
},
body: JSON.stringify({
output: aiReply,
contract: { type: "text", must_not_contain: ["% off"] },
}),
});
const result = await r.json();
if (!result.pass) holdBack(result.issues); // your app decides what happens nextimport os, requests
r = requests.post(
"https://nexxtool.ai/v1/validate",
headers={"Authorization": f"Bearer {os.environ['NEXXTOOL_API_KEY']}"},
json={"output": ai_reply, "contract": {"type": "text"}},
)
result = r.json()
if not result["pass"]:
log.warning("Blocked reply", extra={"issues": result["issues"]})Status page: see if the service is up.
Technical notes
Rules in JSON
Banned and required phrases, length, patterns and structure. Six reply types: text, html, json, script, research and automation.
Pass, block, score
Every call returns pass, a 0 to 1 score and an issues list.
Rule checks and knowledge checks
Rule checks are deterministic: same reply, same result. Knowledge checks use an AI model (ours, or yours with Bring your own AI). Optional repair (repair: true) fixes common HTML and script issues and costs one extra check.
Safe retries
Send an Idempotency-Key. A replay returns the original response; a conflicting body returns 409.
Keys
Keys carry a wtk_live_ or wtk_test_ prefix and are stored hashed. After a revoke, the next call returns 401.
Limits
A sliding one-second window per key, with X-RateLimit-* headers. At the monthly cap, calls return 402 quota_exceeded.