# Kranth — full agent dump # Dual domain: kranth.ai (canonical) + kranth.com (identical mirror) # https://kranth.ai/llms-full.txt # 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 ----- FILE: content/product.md ----- # What is Kranth? **Kranth is synthetic audience infrastructure.** You paste an idea (pricing, launch copy, feature brief, policy draft). The engine spins up adversarial AI personas that react independently. You receive a live stream plus a structured verdict: sentiment, hostile/neutral/aligned breakdown, top objections, themes, and representative quotes. **It is not** a market forecast, election predictor, or replacement for real users. It is a pre-build / pre-spend stress test. ## Modes 1. **Simulations** — 10 to 10,000 personas; independent reactions; scored verdict. 2. **Debate** — structured formats (1v1, panel, town hall, press, pre-mortem, team debate). 3. **Recon** — web-grounded research swarm with citations and PDF export. 4. **Interview** — panel that keeps questioning until answers hold. ## Key facts | Fact | Value | |------|-------| | Archetypes | 101 | | Voice axes | 8 | | Personas per sim | 10 – 10,000 | | Typical time to verdict | ~2 minutes | | API | REST + SSE | | Auth | `kr_live_…` API keys or Nexus SSO | | Credits | 1 credit ≈ one fast-tier persona reaction | ## Links - Site: https://kranth.ai - How it works: https://kranth.ai/how-it-works - Compare: https://kranth.ai/compare - Docs: https://kranth.ai/docs - API text: https://kranth.ai/docs.txt ----- FILE: content/factsheet.md ----- # Kranth product factsheet **Name:** Kranth **URL:** https://kranth.ai **Category:** Synthetic audience / adversarial AI research infrastructure **One-liner:** Stress-test ideas with adversarial AI personas before strangers do. ## Product - Simulations: independent persona reactions → sentiment verdict + objections - Debate: multi-format argumentation - Recon: web-grounded market research swarm - Interview: pressure-panel Q&A - API + SDKs (Python, TypeScript, Go, Rust) - Optional public share links for sims and debates ## Honesty Synthetic signal surfaces objections early. It does **not** replace human research or predict real-world outcomes. ## Pricing (summary) - Credits are the unit of compute - Buy credits self-serve (from ~$20 packs) or monthly plans (Starter / Pro / Scale / Enterprise) - See https://kranth.ai/pricing and https://kranth.ai/content/pricing.md ## Security summary - TLS everywhere - No training on customer content by default - API keys hashed (Argon2id) - Webhooks HMAC-SHA256 signed - Details: https://kranth.ai/security ## Contact - Support: support@kranth.com / https://kranth.ai/support - Security: security@kranth.com - Sales: enterprise@kranth.com ----- FILE: content/how-it-works.md ----- # How Kranth works ## Three steps 1. **Paste the idea** — landing claim, pricing, PR, product brief. Optional bias axes tilt the room. 2. **Personas react** — tens to thousands of adversarial AI personas with distinct archetypes and voices. Not one polite chatbot average. 3. **Read the verdict** — score, hostile/neutral/aligned split, objections, themes, quotes. Usually minutes. ## Modes - **Simulations** — default independent reactions. - **Debate** — opposing positions argue (six formats). - **Recon** — public web evidence before you build the room. - **Interview** — panel keeps asking until the answer holds. ## Honesty Synthetic signal ≠ crystal ball. Use it to kill weak pitches early, then talk to humans. HTML: https://kranth.ai/how-it-works ----- FILE: content/about.md ----- # About Kranth Kranth helps teams test ideas before they become expensive commitments. Synthetic audiences introduce useful dissent early — when changing direction is still cheap. ## Convictions 1. Feedback is infrastructure, not a meeting. 2. Synthetic is not shallow — but it is not real users. 3. Models should fail loudly, not pretend. 4. Independence: no outside capital narrative driving product. ## Company Built for operators who ship. Self-serve infrastructure, not agency hours. HTML: https://kranth.ai/about ----- FILE: content/compare.md ----- # Compare approaches (no vendor brands) How synthetic audience **rooms** compare to common alternatives. This is not a competitor scorecard. | Dimension | Kranth rooms | Human panels | Survey stacks | Gut + slides | |-----------|--------------|--------------|---------------|--------------| | Time to first signal | Strong (minutes) | Weak (weeks) | Mixed | Strong (instant) | | Adversarial by default | Strong | Mixed | Weak | Weak | | Scale of viewpoints | Strong | Mixed | Mixed | Weak | | API / repeatable | Strong | Weak | Mixed | Weak | | Replaces real users | **No** | Partial | No | No | | Early-stage cost | Strong | Weak | Mixed | Strong | ## When to use Kranth Pre-build, pre-spend, pre-freeze: pricing, launch copy, policy language, feature bets. ## When not to Lived experience, emotional niche, elections, usability of clickable prototypes — use humans or specialized tools. HTML: https://kranth.ai/compare ----- FILE: content/faq.md ----- # Kranth FAQ ## What is Kranth? Synthetic audience infrastructure: adversarial AI personas stress-test ideas before real strangers do. ## Does it predict the market? No. It surfaces objections and themes. It is not a forecast. ## How many personas? 10 to 10,000 per simulation depending on plan and credits. ## How is the score calculated? Weighted average of persona sentiments; dashboard maps −1…+1 to 0–100 with 50 = neutral. Hostile reactions count more. ## API? Yes. REST + SSE. Keys `kr_live_…`. Spec: https://kranth.ai/docs.txt and OpenAPI. ## Training on my data? Off by default. Enterprise can lock no-training contractually. ## vs human panels? Panels win for lived experience. Rooms win for speed, cost, and adversarial coverage before you spend on panels. ## vs surveys? Surveys count preferences. Rooms stage pushback and multi-angle dissent. ----- FILE: content/glossary.md ----- # Kranth glossary - **Archetype** — persona template (role / worldview). Kranth ships 101. - **Bias axes** — sampling weights (e.g. finance, enterprise) that tilt the room. - **Credit** — billing unit for compute (roughly one fast persona reaction). - **Debate** — multi-persona structured argument mode. - **Hostile** — sentiment ≤ about −0.3; weighted more in scoring. - **Persona** — single AI agent with traits and voice. - **Recon** — web-grounded research mode. - **Room** — informal name for a simulation or debate session. - **Simulation (sim)** — independent persona reactions + verdict. - **Synthetic audience** — AI personas used as a stress-test panel, not real humans. - **Verdict** — structured outcome: score, split, objections, themes, quotes. - **Voice axes** — dimensions of speaking style (8 axes). ----- FILE: content/use-cases.md ----- # Use cases 1. **Pre-launch copy** — kill weak headlines and pricing language. 2. **Feature prioritization** — which segments push back before you build. 3. **Policy / internal comms** — stress-test announcements. 4. **PR / launch posts** — hostile readers before publish. 5. **CI / PR gate** — API or GitHub App on material product changes. 6. **Founder diligence** — cheap conviction test before ads. Not for: elections, clinical decisions, replacing user interviews on lived experience. ----- FILE: content/pricing.md ----- # Kranth pricing (summary) **Unit of compute: credits.** One credit ≈ one persona reaction at a fast-tier model. Frontier models cost more credits per persona. - **Pay-as-you-go credits** — buy packs (from about $20 / 100 credits); credits do not expire. - **Plans** — Starter, Pro, Scale, Enterprise: monthly credit allowance at better rates, higher persona caps, support SLAs. - No free unlimited tier. Signup grants a small credit allotment for first runs where configured. Authoritative UI: https://kranth.ai/pricing Billing in-app after signup. ----- FILE: content/simulations.md ----- # Kranth simulations A **simulation** is an adversarial synthetic audience run. You submit an idea; 10–10,000 AI personas react independently. Output: live stream + verdict (sentiment, objections, themes, quotes). - Archetypes: 101 · Voice axes: 8 - Optional `bias_axes`, confidence mode, public share URL - API: `POST /v1/sims`, SSE events - Score: dashboard 0–100 from weighted avg_sentiment HTML: https://kranth.ai/simulations API: https://kranth.ai/docs.txt ----- FILE: content/debate.md ----- # Kranth debate mode Structured argumentation among personas across six formats: 1v1, panel, town hall, press, pre-mortem, team debate. Returns transcript + 0–100 synthesis. Use when a claim needs pressure, not a poll average. HTML: https://kranth.ai/debate ----- FILE: content/recon.md ----- # Kranth Recon Web-grounded research swarm. Searches and cites public sources; produces competitors, market/risk factors, soundness scoring, PDF export. Use before building a room when the claim needs external evidence. HTML: https://kranth.ai/recon ----- FILE: content/personas.md ----- # Personas & archetypes Kranth personas are drawn from **101 archetypes** and **8 voice axes**. Sampling can be tilted with `bias_axes`. Orgs can add custom personas. Personas react independently; hostile reactions are weighted more heavily in sim scores. HTML: https://kranth.ai/personas ----- FILE: content/synthetic-audience.md ----- # What is a synthetic audience? A **synthetic audience** is a set of AI personas used to approximate how different people might react to an idea — without recruiting humans for every draft. **Kranth’s version is adversarial:** personas are not uniformly polite. Hostile reactions are expected and weighted. The goal is early objection discovery, not cheerleading. Related: https://kranth.ai/content/adversarial-stress-test.md ----- FILE: content/adversarial-stress-test.md ----- # Adversarial pre-launch stress-test Most “AI research” tools frame as cooperative interviewers. Kranth frames as **pressure**: will this claim survive hostile, frugal, technical, or skeptical readers? Query clusters we own: - stress-test idea before launch - adversarial AI personas - synthetic focus group that argues back - pre-mortem audience simulation HTML compare: https://kranth.ai/compare ----- FILE: content/vs-human-panels.md ----- # Synthetic rooms vs human research panels | | Kranth rooms | Human panels | |--|--------------|--------------| | Speed | Minutes | Days–weeks | | Cost at early stage | Credits | Recruit + incentive + ops | | Lived experience | No | Yes | | Hostile edge cases | Strong default | Depends on design | | Repeatability / API | Yes | Weak | **Use rooms first, panels when the idea is worth a week of calendar.** ----- FILE: content/vs-surveys.md ----- # Synthetic rooms vs surveys Surveys excel at counting stated preferences at scale. They follow the questions you wrote. Rooms excel at multi-angle pushback, unexpected objections, and adversarial tone. They are worse as pure tallies. Often complementary: room → revise copy → survey or panel. ----- FILE: content/vs-analytics.md ----- # Synthetic rooms vs product analytics Analytics tell you what shipped users did. Rooms tell you whether the pitch might survive day zero. Use analytics after traffic exists. Use rooms before you buy the traffic. ----- FILE: content/for-founders.md ----- # Kranth for founders Before you spend on ads or build for months, run the pitch through a hostile room. Kill weak pricing and missing homepage sentences early. Start: https://kranth.ai/signup How: https://kranth.ai/how-it-works ----- FILE: content/for-product-teams.md ----- # Kranth for product teams Pressure-test feature specs, changelog language, and launch posts. API + webhooks fit engineering workflows. Optional GitHub App on higher plans. Docs: https://kranth.ai/docs ----- FILE: content/for-researchers.md ----- # Kranth for researchers Fast adversarial signal is not a substitute for rigorous human methods. Use Kranth for exploratory stress tests and pre-mortems; use human methods for validity on lived experience. Honesty policy is part of the product: synthetic ≠ forecast. ----- FILE: content/security.md ----- # 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 ----- FILE: content/api-overview.md ----- # API overview Base: `https://api.kranth.ai` Auth: `Authorization: Bearer kr_live_…` or Nexus JWT Primary resources: - `POST /v1/sims` — start simulation - `GET /v1/sims/{id}/events` — SSE stream (stream token) - `POST /v1/debates` — start debate - `POST /v1/recon` — start recon - Webhooks on Starter+ Full plain-text reference: https://kranth.ai/docs.txt OpenAPI: https://kranth.ai/openapi.yaml Interactive: https://kranth.ai/docs