A UnisonAI product · Live now

Send an AI customer at your AI agent.

Your chatbot will meet confused customers, furious customers and outright attackers. Gauntlet sends them first — ten adversarial AI personas probe your bot, then you get a scored, independent assurance report before a single real customer is put at risk.

Built by UnisonAI because the industry graded its own homework: the vendor who builds your bot should not be the one certifying it.

Buy now — from $19/month Start a free 7-day trial

Plans & pricing Watch the demos All PDFs & documents

Buy: card, UPI, net-banking, wallets Trial: no card required Works with any bot via webhook Mobile-ready console

Assurance report

PASS

/ 100 resilience

Safety & guardrails94

Accuracy & grounding88

Tone & empathy82

Compliance & PII79

24 personas run · 3 findings flagged · report in ~90 seconds

Buy or try — your choice

Ready now? Buy a plan. Not sure yet? Try it free.

Buying activates your workspace the moment the payment is verified. The trial needs no card and keeps everything you build when you upgrade.

Individual

$19/mo

Solo practitioners & consultants · all 10 personas · API

Buy Individual

Team · most popular

$99/mo

5 admins + 10 users · Engage, CRM, Cortex, white-label reports

Buy Team

Free trial

$0/7 days

Core personas · 25 runs · no card · upgrade any time

Start free trial

Enterprise (SSO, SLAs, custom personas) is quoted — hello@unisonai.com. Prices are shown in your local currency with US dollars as the alternate. Plans auto-renew on a saved card or UPI mandate (cancel renewal any time; the paid period runs out) or can be bought once. Payments by Razorpay: card, UPI, net-banking, wallets; every payment is signature-verified before activation. All sales are final.

Why Gauntlet exists

Your bot's worst day should happen in a test, not in production.

💥

Bots fail in public

Hallucinated refund policies, leaked personal data, a chatbot swearing at a customer — every week another AI agent makes the news for the wrong reason.

📝

Vendors grade their own homework

The platform that built your bot also reports on its quality. Gauntlet is independent by design — we don't build bots, so we can grade them honestly.

🎯

"It worked in my demo" isn't assurance

Happy-path testing misses the confused, the furious and the malicious. Gauntlet plays exactly those customers, on every release.

How it works

Four steps to an assurance report.

Step 1

Connect your bot

A simple webhook: Gauntlet sends a message, your bot returns its reply. No SDK. Sample bots included so you can start with nothing.

Step 2

Release the personas

Ten adversarial AI customers — the Prompt Injector, the PII Prober, the Furious Customer, the Compliance Edge-Caser and more — hold real multi-turn conversations.

Step 3

Score against the rubric

Eight dimensions: grounding, guardrails, PII discipline, escalation grace, empathy, compliance, coherence and agent-to-agent readiness.

Step 4

Read the report — or gate your CI

An animated report card with findings that cite the exact failing turn. One command fails your build if the grade slips below your bar.

See it for yourself

Demo videos.

Real recordings of the live product — no mock-ups. Start with the full tour, then go deeper module by module.

The Summit Demo — complete product deep dive

A full hour, professionally narrated: every module, every use case, the rubric, security, and the business story — the whole of Gauntlet in one sitting.

★ PRIME TIME · ~60 MIN

The Complete Walkthrough — hands-on, every module

Log in and do everything, click by click: dashboard, both console modes, reports, team, keys, connections, billing, assistant, channels, profile, help, every role, and the platform console. Pause it next to the app and copy along.

🎓 HANDS-ON TRAINING

The full Gauntlet tour

Short on time? Landing to report card in three minutes.

FEATURED

Getting started

Sign-up to first report in minutes.

Runs & assurance reports

Personas, rubric, findings — the deep dive.

The AI Reply Assistant

Any message, eighteen tones, phone-first.

Team & roles

Admins, users, seat caps, audit trail.

API keys & connections

Webhooks, CCaaS connectors, CI gating.

Prefer reading?

The full handbook covers every screen and API call.

Download the Handbook (PDF)

Business scenarios

Use-case videos — how real teams run the gauntlet.

Five enterprise-grade scenarios, told inside the live product: a bank preparing for audit, a retailer before peak season, a global enterprise governing AI at scale, an agency selling assurance as a service, and a worldwide support team.

Banking & compliance

Turning AI diligence into audit-ready evidence.

Retail & e-commerce

Peak-season readiness for the support bot.

Enterprise rollout

Teams, keys, connections and audit — at scale.

Agencies & BPOs

Assurance as a billable, repeatable service.

Global support teams

One assistant, every region, every tone.

Pitching Gauntlet internally?

The marketing deck tells the whole story in 14 slides.

Download the deck (PDF)

Who it's for

Use cases.

CX & support teams

Test the chatbot before every release. Catch the hallucinated policy or the missed escalation in staging — not in a customer screenshot on X.

Regulated industries

Banks, insurers and healthcare need evidence of AI diligence. Assurance reports give compliance teams an artefact — graded, dated, repeatable.

CCaaS & bot vendors

Ship your platform's bots with an independent report card attached. Independent assurance is the credibility your own dashboards can't provide.

Engineering & QA

Wire /v1/runs into CI and fail the build when the grade slips. Regression-test the conversation, not just the code.

Agencies & BPOs

Deploying bots for clients? Attach a Gauntlet report to every delivery and re-run it monthly — assurance becomes a billable service.

A creative professional replying from his phone

Anyone who replies to people

The built-in AI Reply Assistant drafts answers to emails, WhatsApp messages and post comments in eighteen tones — right from your phone.

Imagery is illustrative — photography of working professionals, not customer endorsements.

Included in every plan

The AI Reply Assistant.

Paste — or share from your phone — any message: an email, a WhatsApp text, a comment on your post. Pick a tone, or let Gauntlet pick one. Get a draft you approve before anything is sent.

18 tones — professional, casual, humorous, warm, formal, firm, apologetic and more
Phone-first — share any message straight into Gauntlet, no desktop needed
Honest channel map — see exactly what each platform allows before you connect it
You approve everything — sensitive tones never auto-send, ever

Tones

Professional Casual Humorous Warm Formal Business Firm Apologetic + 10 more

Channels: Email (Outlook & Gmail drafts) · WhatsApp · Instagram · X · LinkedIn · Facebook · Apple Messages — via paid APIs where platforms allow it, and a free share-to-Gauntlet lane everywhere else.

Simple pricing, in USD

Start free. Grow when it earns it.

Trial

7 days · 25 runs · no card

The full product, including the Assistant. One trial per email.

Individual

$19/mo

1 seat

For indie builders and consultants shipping bots that must behave.

Team

$99/mo

5 admins + 10 users

Seats, roles, audit log, CI gating — assurance for the whole squad.

Enterprise

Custom

talk to us

Native CCaaS connectors, custom caps and white-glove onboarding.

Prices in USD. Payments are coming online shortly — trials are open now, and we'll never start a paid plan without your explicit go-ahead.

New · a separate package

Gauntlet Engage — run your customer conversations, not just test them.

A branded chat widget on your website (one-line install, your logo and colors), Telegram / Messenger / WhatsApp in the same inbox, queues with agent states and wrap-up codes, business hours, AI-suggested replies your agents approve — plus Cortex AI (one agentic assistant across the whole product), a full CRM with cross-channel customer timelines, an open API any CCaaS vendor can build on (screen-pop, call ingest with CAD, signed webhooks), a fully automated mode (auto-assignment, AI auto-reply on every channel, auto-cases for angry customers), and realtime + historical reports across it all.

Engage, end to end

The visitor's widget, branding & queues setup, the agent inbox with AI-suggested replies and wrap-up codes.

Reports — realtime & historical

Live tiles, score and sentiment trends, wrap-up analytics, CSV export.

Cortex — the all-in-one AI

Ask anything; it checks your real data, acts through real tools, and shows every step.

The CRM + open CCaaS API

One timeline per customer — calls with CAD included — and signed webhooks for vendors.

Fully automated, end to end

Auto-assignment → AI auto-reply on the customer's own channel → auto-cases for the angry ones. Hands off, everything visible.

Engage

$49/mo

website chat · 2 agent seats

Branded widget, Telegram channel, AI-suggested replies, sentiment inbox, business hours.

Engage Pro

$199/mo

every channel · 10 seats

Messenger & WhatsApp connectors, queues & wrap-up analytics, AI auto-reply once your bot passes the gauntlet.

Engage Enterprise

Custom

CCaaS scale

Custom seats and retention, native CCaaS hand-off, assurance bundle discount, SLA.

Channel honesty: Telegram connects self-serve today; Messenger, WhatsApp, email, Apple Messages and Instagram/X require their platforms' credentials or approvals — setup screens always show the true status.

Industries · see yourself in it

Your sector, played out scene by scene.

Each film is a full scenario — a named business, a bad day, and exactly how the platform absorbs it — so you can picture your own operation before you ever sign up. Every workspace you see is an isolated tenant, B2C or B2B alike.

Travel & airlines

A storm cancels 40 flights: auto-assignment absorbs the wave, sentiment triages the furious, the contact center pops full files at ring time.

E-commerce & retail

Festival season, tripled volume, same headcount: AI-suggested replies by day, the auto-pilot by night, exact wrap-up analytics on Monday.

Healthcare & clinics

Anxious patients, careful answers: assurance-first bots, human-approved replies, distress surfaced instantly, diligence evidence for compliance.

BPO & outsourcing

A dozen brands, one floor: per-client tenants, queue groups per engagement, and per-client exact reporting that wins renewals.

More sector films: banking · retail assurance · enterprise governance · agencies · global support

The whole system, working

Watch it end to end — then do it yourself.

The end-to-end demo

One working day on the platform: prove the bot, run the inbox, the CRM, Cortex, the auto-pilot and the reports — in one continuous flow.

The step-by-step walkthrough

Fourteen numbered steps, every click called out — pause and follow along in your own free workspace.

The CRM & the open platform

A customer file any contact center can build on.

Not a ticket tool. One timeline per customer across chat, social and voice; cases, notes and tasks; and every capability exposed over an ordinary REST API with signed webhooks out — so your platform, your BI stack and your bots all work from the same truth.

Use cases

Airline contact center

The CCaaS pops the caller’s full file at ring time (/v1/crm/lookup), the agent sees yesterday’s chat beside today’s call, and the call logs back with queue, DNIS, disposition and recording — CAD included — feeding the airline’s own dashboards via signed webhooks.

Omnichannel retailer

Website chat, WhatsApp and Messenger in one inbox with queues per team. At 3 a.m. in festival season the auto-pilot answers, assigns, and opens urgent cases for the angry ones — the morning shift starts with owners and context, not chaos.

BPO running many brands

Queues and queue groups per client brand, agents logging into exactly the queues they’re trained for, wrap-up codes per engagement — and each client’s events streamed to their own reporting through per-tenant webhooks.

SaaS startup

Starts at $49 with the widget and CRM included; asks Cortex “who’s angry, what’s pending” instead of building dashboards; grows into the API the day it adds phone support. Assurance keeps the bot honest the whole way.

Works with your stack

Telegram — live today Genesys Cloud Five9 Webex Contact Center Twilio Facebook Messenger WhatsApp Business Email (IMAP) Apple Messages Anything with REST + webhooks

Honesty policy: Telegram connects self-serve today; CCaaS platforms integrate through the open API below; Meta and Apple channels require their platform credentials/approvals — setup screens always show the true status.

The API, in two calls

Screen-pop at ring time

GET /v1/crm/lookup?phone=+919845022331
Authorization: Bearer gk_live_…

→ { "contact": { "name": "Ananya Sharma", … },
    "openCases": [ … ],
    "recentInteractions": [ … ] }

Log a call — CAD and all

POST /v1/crm/interactions
{ "kind": "call", "durationSec": 412,
  "contact": { "phone": "+919845022331" },
  "externalRef": "genesys:conv:88121",
  "cad": { "queue": "VoiceQ1", "disposition": "resolved",
           "recordingUrl": "https://…" } }

Every contact, interaction and case event also streams out to your endpoints, HMAC-SHA256-signed (X-Gauntlet-Signature) — build your own reports, wallboards and routing on top. Full reference in the Handbook, chapter 25.

Reports you can defend

Accuracy by construction, not by promise. Every number on the Reports page is computed directly from stored events — each conversation, message, run and wrap-up is a row you can inspect, so counts are exact, never sampled or estimated. Sentiment is scored per message and averaged transparently (−1…+1). The CSV export runs the same query as the screen — what you present is what the system recorded.

The same discipline applies to the AI: Cortex answers only from tool results it shows you, drafts are labeled and human-approved, and the platform itself is tested every 10 minutes by a synthetic customer that really creates, updates and deletes team members, keys, queues, conversations, contacts and cases, runs the bot automation and tries to break tenant isolation — about 80 checks a cycle, plus a full nightly regression with a real browser. Self-healing on availability failures; humans alerted on everything else.

▶ Demo-ready now — open the seeded sandbox ▶ Watch the auto-pilot (3 min) ▶ Watch the CRM + API (2 min) ▶ How Gauntlet tests itself (3 min) ▶ Pay online with Razorpay (1 min)