Your data stays yours. Your people stay in charge.
How Kinnow protects data and keeps AI accountable. We list only what is built, and say so when something is still rolling out.
Your data never trains outside AI models.
Your data is never used to train outside AI models. Kinnow's own Instinct models learn from your company's data only if your admin turns that on. It can be switched off at any time. Veil, our privacy layer, holds back bank, tax and ID numbers and swaps names for placeholders before an AI call. It covers Ask Kinnow and most agents today; the rest are being added. Negotiated terms stay home. Every call is recorded: what was sent, what was withheld and why. Answers come back with the real names put back inside Kinnow.
- Masking is rolling out agent by agent; the register shows which calls were masked
- A “what was sent” link on every AI answer
- An exportable register of every AI call for your auditors
Nightly, with faster tiers
Hourly or continuous replication if your business needs it.
Your data is yours
Export any report or register to Excel or CSV whenever you like.
Run all of it inside your own country.
The application, the database, your documents and search can all run within your country's borders, in line with India's DPDP Act and local data-protection law. Each company decides whether AI calls may leave its country. This works today on Ask Kinnow and document search, and is being added to agents. Ask Kinnow has a “keep in country” switch on every question.
- Hosting in your country, chosen at onboarding
- Data location kept for files, search and the AI's index of your documents
- Erasure requests reach files, search and the AI's index of your documents
- Consent recorded per channel for marketing, with the strictest rule applied across countries
Your data, your control.
What is in place today, and what is still rolling out, said plainly. During evaluation, our team takes your IT and audit people through the architecture, controls and audit status in detail.
Six commitments behind every AI action.
Agents prepare. People authorise. You can see why, and you can take it back.
Can Kinnow's AI see all our data? No. Every agent and every Ask Kinnow answer is limited to what the person asking is allowed to see, across structured records and documents. The knowledge layer over email and documents is off until you switch it on.
Data protection
- Your company's data stays separate on every request: each call says which company, business line and branch it is for, the server checks the person belongs there and refuses otherwise, and we test this separation on every code change before it ships.
- Access follows role: agents and Ask Kinnow work only from what the person asking may see, across records and documents (Vault). A mailing list or an answer can never include records its author couldn't see.
- Access is role-based: people see only the screens and data their role allows, every request is checked, and sign-in uses secure, short-lived passes.
- The knowledge layer that reads email and documents (Org-Intelligence) is off by default; your company switches it on. Employees can opt out of having personal email indexed, and all participants are told before a meeting transcript is indexed.
- Email follows role: sales sees their own and their customers' email; accounts sees finance messages only; the super-admin 'see all' view is behind a switch that is off by default.
- Connected email, Drive and meeting data (Signals) is fetched one record at a time when asked, summarised, and the message text is dropped after a 24-hour hold; every outside access is logged and admins can cut off access to any record instantly.
- Records kept: financial records are kept as the law requires (a country's setup may extend this); the email index is kept for a period you set; deleted source emails are removed from the index within 30 days.
- The right to erasure (DPDP / GDPR Art-17) reaches the file, the search index and the AI's index of your documents, without breaking the duty to keep financial records.
- Sensitive document types (legal, board, HR) are private by default, and each type can be kept out of indexing.
- Audit logs cannot be edited or deleted in the application; the only permitted change is replacing a person's name for GDPR (pseudonymisation), and that change is itself logged.
- An optional tamper-evidence check on the audit log (a SHA-256 hash chain, one per company) makes an edit made outside Kinnow detectable: tamper-evident, not tamper-proof. It is off by default and is switched on with an engineer's help today.
- You pick where each company's data lives (with an extra sign-in check, and logged), and it is honoured for stored files, search and the AI's index of your documents. An India-hosted setup sends AI work to an in-region AI provider; the check on each individual AI call is still rolling out (partial).
- Files are stored encrypted with managed keys (KMS, S3 SSE-KMS) and versioned; duplicates are removed within a company only, never across companies.
- Storage never silently deleted: soft cap at zero balance, then 30-day read-only, then 90-day freeze.
- Business continuity: Backup is included; Assured and Critical are optional paid continuity tiers. Restoring a company's snapshot is built.
- Connected content is scanned for hidden instructions aimed at the AI (InjectionGuard), and card details on payments are replaced with tokens under RBI rules.
AI you can account for
- People decide: agents advise and draft; suggestions they are unsure about are only shown, never pre-filled or applied automatically.
- Confidence bands decide what happens: 0.75 and above, 0.60-0.74, below 0.60, each showing its reasons and never a bare score; results from fixed rules are labelled apart from the AI's own inferences.
- Actions that cannot be undone or carry high value (filings, payments, journal posting) are never approved automatically; the higher the stakes, the more checks, up to an extra sign-in check or a second approver.
- The audit footer records who did what, whether a person or an agent, and when, with which model and for which company; regulated companies see 'agent-assisted' labels.
- Your corrections are kept as feedback and used to adjust how confident agents are.
- If agents are down or credits run out, the product stays fully usable by hand. While a company is closing its period, agents are not switched off.
- Cost limits and credit budgets stop a runaway agent from using up your credits; usage alerts arrive before any hard stop.
Close checklist · 80% pre-filled
| Bank reconciliation · 3 accounts | Done |
| Receivables ageing reviewed | Done |
| Accruals: rent, power | Needs you |
| Fixed-asset depreciation posted | Done |
| Intercompany balances tie | Agent checking |
Regulations
What the platform enforces today. We publish independent audit reports here when they are issued.
| Item | Status |
|---|---|
| DPDP Act (India) and GDPR erasure rights | Built in: erasure reaches files, search and the AI's index of your documents |
| RBI card-tokenisation rules | Built into the payment gateway |
| ZATCA Fatoorah (Saudi Arabia) | Planned after launch |
| Financial-record retention as the law requires | Built in; a country's setup can extend it |
Processes and controls you can hand over.
Every ready-made industry setup (we call it an industry pack) comes with its processes: the process, the owner of each step, which steps are enforced by an approval or a limit you set, and which are notes. Print it or create a written procedure from it.
- Month-end close: twelve checks, seven must pass, and only the Controller can reopen a closed month
- Access request to grant, master-data change approval and workflow change to live, in every industry setup
- A stricter approval rule can run as a trial first and report whom it would have stopped
Tell Kinnow what slows you down.
Answer three questions. Get the applications, the agents and a two-week plan for your business.