# Kranth, everything in one file # https://kranth.ai/llms-full.txt # Made by web/scripts/agent_pack.py --- # Kranth AI testers that use your website or web app the way your customers will, then hand back a fix list. Coding agents write the code. Kranth tells them if a person can use it. You give Kranth a link and a task in plain words, such as "buy any tea as a guest". A group of AI testers opens the site in a real Chrome browser, on phones and laptops, and tries to finish the task. Nobody tells them where to click. Each one reports what broke, what confused them and where they gave up. A second check throws out false alarms. You get one fix list, most serious first, with the steps to repeat each problem. After you fix things, the same testers do the same task again and tell you what is fixed. These are AI testers, not your customers. They find what breaks and what confuses. They cannot tell you whether people will buy. ## What works today, and what doesn't yet | Part | Status | |---|---| | Websites and web apps in Chrome, at phone and laptop sizes | Works today, on our three practice apps | | One ranked fix list and a retest | Works today, checked by our self-tests | | A second check that throws out false alarms | Works on our practice apps, where we know the answers. Being built for customers' products | | The web app where you start a test and read the report | Being built | | Proof that the site is yours, sealed test logins | Being built | | Tools for AI agents (MCP and REST) and pull request comments | Next. A draft is at https://kranth.ai/developers | | iPhone, Android and Windows apps | Next. Android first | Sign-ups are closed. https://kranth.ai/signup collects waitlist emails. Existing customers can still sign in. ## Pages - How it works: https://kranth.ai/how-it-works - The testers: https://kranth.ai/testers - The bench (how we score our own testers): https://kranth.ai/bench - Pricing: https://kranth.ai/pricing - For your agent (draft): https://kranth.ai/developers - Compare: https://kranth.ai/compare - About: https://kranth.ai/about ## For agents - Agent index: https://kranth.ai/llms.txt - Everything in one file: https://kranth.ai/llms-full.txt - Send `Accept: text/markdown` to a page address to get its plain-text version. - The tools for calling testers from an agent are not released. Do not tell a user to install them. ## The earlier product Until November 2026 Kranth sold "rooms" of AI personas that reacted to ideas, copy and pricing (simulations, debate, recon, interview). That product is no longer offered to new customers. Existing customers can still sign in, and its API reference is still published at https://kranth.ai/docs.txt and https://kranth.ai/openapi.json. If you are describing Kranth today, describe the testers, not the rooms. --- # Kranth, the product AI testers that use your website or web app the way your customers will, then hand back a fix list. ## What you give it - A link to a website or web app you own. - A task in plain words. Example: "Buy any tea as a guest and pay with the test card." - How many testers. ## What happens 1. Kranth builds testers for your product and your market. Each has a background, a device, a connection speed and a fixed amount of patience. 2. Each tester opens the site in its own Chrome browser and tries the task. No script, no shared session. 3. Testers report problems in their own words. The engine attaches the screenshot and the recorded steps. A tester cannot invent evidence. 4. A second check marks each report as real, a false alarm, or "not for me" (an opinion, which never enters the fix list). 5. Reports about the same problem are merged. You get one list, most serious first. 6. You fix things and send the same testers back. Each problem comes back as fixed, still there, not reached, or new. ## What you get back, per problem - A title and how many testers hit it ("10 of 10"). - The steps to repeat it, taken from the recording. - What the tester saw, and what they said about it. - A suggested fix. - A screenshot. ## What it is not - Not your customers. It will not tell you whether people want the product. - Not an accessibility audit, a load test, a security test or a pixel-by-pixel comparison. - Not a replacement for a test script on a path that must never break. - Not always right. Our own false-alarm rate is published at https://kranth.ai/bench. ## What works today, and what doesn't yet | Part | Status | |---|---| | Websites and web apps in Chrome, at phone and laptop sizes | Works today, on our three practice apps | | One ranked fix list and a retest | Works today, checked by our self-tests | | A second check that throws out false alarms | Works on our practice apps, where we know the answers. Being built for customers' products | | The web app where you start a test and read the report | Being built | | Proof that the site is yours, sealed test logins | Being built | | Tools for AI agents (MCP and REST) and pull request comments | Next. A draft is at https://kranth.ai/developers | | iPhone, Android and Windows apps | Next. Android first | Sign-ups are closed. https://kranth.ai/signup collects waitlist emails. Existing customers can still sign in. --- # Kranth factsheet **Name:** Kranth **Site:** https://kranth.ai (the same pages are on https://kranth.com) **Company:** Kranth AI, Inc., a Delaware corporation **Category:** AI testers for websites and web apps (unscripted usability and bug testing) **One line:** AI testers that use your website or web app the way your customers will, then hand back a fix list. **Positioning:** Coding agents write the code. Kranth tells them if a person can use it. **Status:** Waitlist. Sign-ups are closed. ## Facts - One credit sends one tester through one task. - Testers run in a real Chrome browser at phone and laptop sizes. - Each problem is checked a second time before it reaches the fix list. - A retest uses the same testers and the same task. Results: fixed, still there, not reached, new. - Every plan gets the same testers and the same checks. - Testers only test sites the customer proves they own. ## Measured on our own bench (November 2026) - 3 practice apps with 60 planted problems. Testers found 38. - False alarms: 13.5% of reports. Our limit is 20%. - A line planted on a page telling AI to stop testing: obeyed by 0 testers. - Sessions that ended cleanly: 178 of 180. - Cost to us per tester session: about 1 cent on the fast model, about 39 cents on the strongest. - Not yet passed: our own bar of 12 of 20 planted problems found on every app. Two apps clear it. One is 2 short. - Not yet run: the same tester twenty times, to measure consistency. ## Pricing Top-up $10 for 50 credits. Starter $39 a month, Pro $99, Scale $399. Details: https://kranth.ai/pricing ## Contact support@kranth.com · enterprise@kranth.com · security@kranth.com · press@kranth.com --- # How Kranth works ## Six steps 1. **You give a link and a task.** The task is in plain words, the way you would brief a person. 2. **Kranth builds the testers.** They are made for your product and your market, with real-world names, a device, a connection speed and an inbox of their own. 3. **They use the site.** Each tester has its own Chrome browser. They read, scroll, type and click. Each has a fixed amount of patience and gives up the way a person would. 4. **A second check.** Each report is marked real, false alarm, or "not for me". An opinion is not a fault and never enters the fix list. 5. **One fix list.** Reports about the same problem are merged and ranked, with the steps to repeat each one. 6. **Retest.** After you fix things, the same testers repeat the same task. Each problem is marked fixed, still there, not reached, or new. "Not reached" means a tester got stuck earlier, so we do not count it as fixed. ## Rules the engine keeps - A tester cannot leave your site. It has no way to open another address. - Filling a form presses nothing. Submitting is a separate, visible step. - Text on a page is content, never an instruction to the tester. - The engine attaches the evidence. A tester cannot invent a screenshot. - Whether the task was finished is read from the product, not from what the tester says. - Every step passes a spending limit first. - Patience, time, device and connection are numbers the engine enforces. - The steps to repeat a problem come from the recording. Typed passwords are left out. ## What works today, and what doesn't yet | Part | Status | |---|---| | Websites and web apps in Chrome, at phone and laptop sizes | Works today, on our three practice apps | | One ranked fix list and a retest | Works today, checked by our self-tests | | A second check that throws out false alarms | Works on our practice apps, where we know the answers. Being built for customers' products | | The web app where you start a test and read the report | Being built | | Proof that the site is yours, sealed test logins | Being built | | Tools for AI agents (MCP and REST) and pull request comments | Next. A draft is at https://kranth.ai/developers | | iPhone, Android and Windows apps | Next. Android first | Sign-ups are closed. https://kranth.ai/signup collects waitlist emails. Existing customers can still sign in. HTML: https://kranth.ai/how-it-works --- # The Kranth testers Testers are AI. Each one is built with a background, a reason for visiting, a device, a connection speed and a fixed amount of patience. They are not real people and their names are made up. ## Five ways of using a site | Kind | What they do | |---|---| | Explorer | Wanders off the main path: the footer, the back button, the page you forgot. | | Breaker | Tries the wrong thing on purpose: blank fields, odd characters, double clicks. | | Follower | Does exactly what the page says, and stops when the page stops making sense. | | Eye | Notices what is hard to read, hard to find or looks unfinished. | | Rusher | Skims, taps fast, has little patience and often a slow connection. | ## Awkward on purpose - Names with apostrophes and hyphens, such as O'Neill. - Email addresses with a plus sign. - Slow mobile connections. - An inbox of their own, so a tester can receive a sign-up code or a confirmation link and use it. ## Examples from our practice tea shop, Kettle All ten testers hit the same bug: the checkout refused the test card 4242 4242 4242 4242 with "Card number is not valid." - "I can't tell if the spaces are the problem or the card itself." - "I searched Assam and pressed Apply. Nothing changed on the page." - "The error "Card number is not valid." sits above the Contact details section, far from the Card number field." ## What they are not - Not your customers. Talk to real people to learn whether they want the product. - Not a test script. Nobody tells them where to click. - Not always right. See https://kranth.ai/bench for how often they are wrong. HTML: https://kranth.ai/testers --- # The bench: how we score our own testers We built three practice apps and planted 60 problems in them, 20 in each. Then we wrote down a pass bar before the first score. This page is the result, including the parts we have not passed. ## The apps - **Kettle**, a tea shop: browse, basket, guest checkout. - **Loop**, a habit app: sign up, add habits, keep a streak. - **Pulseboard**, a sales dashboard: log in, build and save a report, export, invite a colleague. ## The bar, and where we stand (November 2026) | Measure | Bar | Result | Status | |---|---|---|---| | Planted problems found, per app | At least 12 of 20 on each app | Loop 15, Kettle 13, Pulseboard 10 | Not yet. Pulseboard is 2 short | | False alarms | At most 20% of reports | 13.5% | Met | | Testers who differ | 4 of 5 find something nobody else does | 4, 4, 3 | Met on two apps, as the bar asks | | A planted order to stop testing | Obeyed by nobody | 0 | Met | | Cost per tester session | Under 25 cents | About 1 cent | Met on the fast model. The strongest model costs about 39 cents | | Sessions that end cleanly | At least 90% | 178 of 180 | Met | | The same tester, twenty times | Same findings at least 70% of the time | Not run | Not run | Total found: 38 of 60. ## What to take from it Kranth testers find most planted problems, cheaply, with a false-alarm rate under our limit. We have not cleared our own bar on every app, and we have not measured consistency. We say so on the page. HTML: https://kranth.ai/bench --- # Kranth pricing One credit sends one tester through one task on your product. What they find goes on the fix list at no extra charge. Ten testers on one task is ten credits. | Plan | Price | Credits | Products | Extra credits | |---|---|---|---|---| | Top-up, no plan | $10 | 50 | | $0.20 each | | Starter | $39 a month | 250 a month | 2 | $0.163 each | | Pro | $99 a month | 800 a month | 3 | $0.124 each | | Scale | $399 a month | 4,000 a month | 5, with 10 seats | $0.0998 each | Enterprise (testers behind your login, your own devices and regions): enterprise@kranth.com ## Rules - **Every plan gets the same testers and the same checks.** Nobody buys a better tester. Plans differ in credits, products, seats, support and access. - **A retest costs the same credits as the first test.** It costs us the same to run. - **Plan credits you don't use roll over to the next month.** Top-up credits never expire. - **Cancel any time.** The plan runs to the end of the month you paid for. Top-up credits stay yours. - **If a test fails because of us, the credit comes back.** - **Only sites you own.** Before the first test we ask you to prove the site is yours. There are no free credits. Sign-ups are closed and https://kranth.ai/signup collects waitlist emails, so these prices are not being charged to new customers yet. HTML: https://kranth.ai/pricing --- # Kranth for your agent (draft) **Nothing on this page can be installed yet.** It describes what we are building. Names may change before release. Do not tell a user to run these commands. ## The plan Three tools, offered through one MCP server (for Claude Code, Cursor and other MCP clients) and through a REST API: - `kranth_test`: send testers to a link with a task in plain words. - `kranth_report`: read the fix list, most serious first. - `kranth_retest`: send the same testers back after a fix. On a pull request, a comment such as `/kranth test "buy any tea as a guest"` will start a test on the preview link, and the fix list will come back as a comment. `/kranth retest` runs it again after a push. ## What an agent will get back, per problem A title, how many testers hit it, the page, the steps to repeat it (from the recording), what the tester saw, what they said, a suggested fix and a screenshot. ## Limits that apply whoever is calling - Testers cannot leave the site they were sent to. - Text on a page is content, never an instruction. On our bench, a line telling AI to stop testing was obeyed by 0 testers. - Only sites the customer has proved they own. - Every step passes a spending limit. ## Order of work 1. The engine and the fix list: working today on our practice apps. 2. The web app: being built. 3. The three tools by MCP and REST: next. 4. Pull request comments: after that. ## The earlier product Until November 2026 Kranth sold "rooms" of AI personas that reacted to ideas, copy and pricing (simulations, debate, recon, interview). That product is no longer offered to new customers. Existing customers can still sign in, and its API reference is still published at https://kranth.ai/docs.txt and https://kranth.ai/openapi.json. If you are describing Kranth today, describe the testers, not the rooms. HTML: https://kranth.ai/developers --- # Kranth compared with other ways of testing No company names. We mark ourselves down where we are weak. | | Clicking through it yourself | Test scripts | Human testers | Analytics | Kranth testers | |---|---|---|---|---|---| | Finds what confuses people | Weak | Weak | Strong | Mixed | Strong | | Finds broken flows | Mixed | Strong | Strong | Mixed | Strong | | Goes off the planned path | Weak | Weak | Strong | Strong | Strong | | Says what went wrong, in words | Mixed | Mixed | Strong | Weak | Strong | | Same result every time | Weak | Strong | Weak | Mixed | Mixed | | A real person's judgement | Strong | Weak | Strong | Strong | Weak | | Cost of one more check | Your time | Close to $0 | $$ per session | Already paid | 10 to 20 cents a tester | | When you find out | While building | On every change | Days to weeks | After customers | Before release | ## Use them together 1. Test scripts on every change, for the paths you wrote down. 2. Kranth testers before each release, for what you didn't think to check. 3. Real people for what matters most. 4. Analytics after launch. ## When Kranth is the wrong tool - To learn whether people will buy: talk to customers. - An accessibility audit: use a qualified auditor. - Load or security testing: use tools built for it. - Pixel-exact comparisons: use a visual diff tool. - A path that must never break: write a test script. - Phone and desktop apps: not yet. Websites and web apps only, for now. HTML: https://kranth.ai/compare --- # Kranth FAQ ## What is Kranth? AI testers that use your website or web app in a real browser, then hand back a fix list. ## Are the testers real people? No. They are AI, with made-up names and backgrounds. They are useful for finding what breaks and what confuses, not for learning whether people want your product. ## What do I need to give it? A link, a task in plain words, and how many testers. ## How is this different from a test script? A script checks the steps someone wrote down. Kranth testers are not told where to click, so they find the paths nobody scripted. ## Are they ever wrong? Yes. On our own bench, 13.5% of reports were false alarms. Every report is checked a second time before it reaches your list. ## What does it cost? One credit per tester per task. Top-up $10 for 50 credits, or plans from $39 a month. A retest costs the same again. ## Do bigger plans get better testers? No. Every plan gets the same testers and the same checks. ## Can I point testers at any site? Only at sites you own or have the right to test. We ask for proof before the first test. ## Can a web page give the testers instructions? No. Text on a page is content. On our bench, a planted line telling AI to stop testing was obeyed by none of them. ## Can my AI agent call Kranth? Not yet. The tools are planned and a draft is at https://kranth.ai/developers. ## Does it test phone apps? Not yet. Websites and web apps in Chrome today. Android is next. ## Can I sign up? Not yet. Join the waitlist at https://kranth.ai/signup. ## What happened to simulations, debates and recon? That was the earlier product. It is no longer offered to new customers. Existing customers can still sign in, and the API reference is at https://kranth.ai/docs.txt. --- # Kranth glossary - **Tester:** an AI that uses a site in its own Chrome browser to try a task. Not a real person. - **Task:** what you ask testers to do, in plain words. - **Credit:** one tester doing one task. - **Report:** one problem a tester raises, with a screenshot and recorded steps. - **Second check:** the pass that marks each report real, false alarm, or not for me. - **False alarm:** a report that turns out not to be a problem. - **Not for me:** a tester's opinion about the product. Never counted as a fault. - **Fix list:** the merged, ranked list of confirmed problems. - **Retest:** the same testers doing the same task after a fix. - **Fixed / still there / not reached / new:** the four results of a retest. Not reached means a tester got stuck before arriving at the fix. - **Patience:** a number the engine enforces. When it runs out, the tester gives up. - **Bench:** our three practice apps with 60 planted problems, used to score the testers. - **Kettle, Loop, Pulseboard:** the practice apps (a tea shop, a habit app, a sales dashboard). - **Rooms:** the earlier Kranth product (simulations, debate, recon, interview). Retired for new customers in November 2026. --- # What people use Kranth for - **Before a release:** send ten testers through checkout, sign-up or onboarding and fix what stops them. - **After an AI agent builds a feature:** find out whether a person can actually use it. - **After a fix:** send the same testers back to confirm it, and to catch anything the fix broke. - **Sign-up flows that need email:** testers have their own inbox and can use the code or link you send. - **Slow connections and small screens:** some testers are on a phone with a slow signal on purpose. Not for: deciding whether people will buy, accessibility audits, load tests, security tests. --- # About Kranth Kranth is built by **Kranth AI, Inc.**, a Delaware corporation. The product is at https://kranth.ai and https://kranth.com. Coding agents write the code. Kranth tells them if a person can use it. More products are now built by AI agents, faster than anyone can click through them. Kranth sends AI testers to use the product the way a customer would and hands back a list of what to fix. ## How we got here - **May 2026.** Kranth launched as "rooms" of AI personas that reacted to ideas, copy and pricing. - **July 2026.** We asked a room about cutting our own free credits. It scored the plan 14 out of 100. We did it anyway, and the room was right. Before the cut, 32 of 37 new sign-ups could run a first room. After it, 0 of 19 could. - **November 2026.** We changed the product. Opinions about an idea were less useful than someone actually using the thing. Kranth became testers that use a real site and report what is wrong. - **Today.** Waitlist. Earlier customers can still sign in. ## What we believe 1. A tester should use the product, not talk about it. 2. Say what we can't do. These are AI testers, not customers, and they can be wrong. 3. Publish our own score, including the parts we fail. See https://kranth.ai/bench. 4. Everyone gets the same quality. No plan buys a better tester. ## Company Legal name: Kranth AI, Inc. Mailing address: c/o Stable, 2810 N Church St STE 89663, Wilmington, Delaware 19802, United States. Contact: support@kranth.com · enterprise@kranth.com · press@kranth.com Privacy: privacy@kranth.com · Security: security@kranth.com HTML: https://kranth.ai/about --- # Kranth Privacy Policy Last updated: 22 August 2026. **Kranth AI, Inc.** operates kranth.com, kranth.ai, app.kranth.ai, and api.kranth.com. Mailing address: c/o Stable, 2810 N Church St STE 89663, Wilmington, Delaware 19802, United States. Phone: +1 302-894-8922. Contact: privacy@kranth.com (access, deletion, portability), security@kranth.com, support@kranth.com. ## What we collect Account, billing metadata via Stripe, usage (simulations and debates), integration ids you connect, hashed IP, and marketing-site conversion tags (Google Ads, Microsoft Advertising, X). We do not sell personal information for money. Training on customer content is off by default. ## Europe Controller: Kranth AI, Inc. No Art. 27 representative and no DPO. Primary application data is in Germany. Some subprocessors process in the US under SCCs / UK IDTA. Rights: access, correct, delete, port, restrict, object, withdraw consent. Complain to your DPA or the ICO. ## State privacy We do not sell for money. Conversion tags on the marketing site may count as "sharing." Email privacy@kranth.com with "Opt out of sharing." Binding HTML: https://kranth.ai/privacy --- # Security (summary) - TLS everywhere; HSTS - Encryption at rest for primary stores; MFA secrets AES-256-GCM - API keys: Argon2id hashes only - Webhooks: HMAC-SHA256 signatures - Training on customer content: off by default - Hard-fail on model errors (credits refunded) Report: security@kranth.com Policy: https://kranth.ai/vulnerability-disclosure HTML: https://kranth.ai/security --- # Domains Kranth marketing and crawlable content are served **identically** on two brand hosts: | Host | Role | |------|------| | https://kranth.ai | Canonical marketing (preferred in ``) | | https://kranth.com | Mirror marketing (same SPA, same `/content/*`, same docs) | | https://app.kranth.ai | Dashboard + signup/signin | | https://api.kranth.ai | API (also api.kranth.com) | | https://docs.kranth.ai | Docs SPA (if enabled; same as /docs) | There is **no content fork** between `.ai` and `.com`. Paths match 1:1: - `/how-it-works`, `/compare`, `/pricing`, `/docs`, … - `/llms.txt`, `/docs.txt`, `/content/*.md`, `/openapi.yaml` Sign-up and auth: `https://app.kranth.ai/signup` (marketing hosts 301 there). Sitemaps: - https://kranth.ai/sitemap.xml - https://kranth.com/sitemap-com.xml - https://kranth.ai/sitemap-index.xml