FinAIAboutContactSupportPrivacySign InBook a Demo →
FinAI · Secure Enterprise AI · New

Meet FinAI. One secure door to every frontier model.

Your finance team already wants ChatGPT, Claude, and Gemini. FinAI gives them all of it in one chat window your organisation controls: every prompt is screened and masked before it leaves your network, the best model answers it, the response is checked on the way back, and the whole exchange is recorded in a tamper-evident ledger. The intelligence gets in. Your sensitive data doesn't get out.

One governed chat overClaudeGPT-classGeminiMistral
THE PROBLEM

The Same Question, Two Very Different Outcomes

Right now, someone on your team is pasting client data into a personal ChatGPT tab — because the tools are genuinely useful and there is no sanctioned alternative. FinAI is that alternative. Watch what happens to the exact same prompt on each path.

Today · Without FinAI

An employee pastes it into a public chatbot

chat.public-ai.com — personal account
Summarise the NPA provision for Vertex Finance Ltd · PAN AAACV9012F
Sent raw to a public consumer service
Client name and PAN exposed in full
?Retention & training use — unknown to you
No record your organisation can audit
Your most sensitive numbers now live somewhere you can't see, delete, or prove.
With FinAI

The same question, through the governed gateway

finai.finifrs.com — your tenant, SSO
Summarise the NPA provision for Vertex Finance Ltd · PAN AAACV9012F
PII masked before it leaves your network
Routed to the best frontier model · zero retention
Response screened before anyone sees it
Hash-chained audit record written
Same intelligence, same speed — and the sensitive data never left your control.
Same frontier intelligence either way —— only one path is governed.
WHAT FINAI DOES
PRINCIPLE · 01

One Chat, Every Model

Your team gets a single, familiar chat window instead of four different public AI tabs. Behind it, FinAI picks the best frontier model — Claude, GPT-class, Gemini, or Mistral — for each question. We broker the models; we never train our own on your data.

PRINCIPLE · 02

Sensitive Data Never Leaves Raw

Before a single token leaves your organization, the prompt firewall finds PII — PAN, Aadhaar, client names — and secrets, then masks or blocks them. The model only ever sees the masked version. Nothing is silently forwarded.

PRINCIPLE · 03

Every Call Is on the Record

Who asked, when, which model answered, what it cost — every call is written to an append-only, hash-chained ledger. A daily integrity sweep flags any tampering before an auditor would ever need to ask.

HOW IT WORKS

Every Request Passes Through One Governed Path

Five stops, every time. Your teammate asks a question like they would in any AI chat. The firewall masks anything sensitive. The router hands the masked prompt to the best-fit model under zero data retention. The answer is screened on the way back. And the whole exchange is sealed into an immutable ledger — before it ever reaches a screen. The trace below is a live simulation of that exact pipeline.

finai_secure_gateway · live request trace
GOVERNANCE ACTIVE
What the employee typed — and what actually leaves your network
Draft the board note on ECL movement for Vertex Finance Ltd · PAN AAACV9012F vs prior quarter.
⚠ 2 sensitive fields detected — client name, PAN
01 · ask
Employee prompt
typing…
your teammate asks, like any AI chat
02 · screen
Prompt firewall
standby
PII & secrets are masked right here
03 · route
Router → Claude
standby
best frontier model, masked prompt only
04 · check
Response governance
standby
the answer is screened on the way back
05 · record
Audit ledger
hash-chained
tamper-evident proof of every call
One gateway, every frontier model:ClaudeGPT-classGeminiMistral+ more, per policy
THE ARCHITECTURE

Ten Controls, One Choke Point

SSO / MFA authentication with DB-driven, role-based access control
Row-level tenant isolation — no organization ever sees another's data
Prompt firewall: PII detection & redaction before a request is sent
Secret scanning — API keys and credentials in a prompt are blocked, not forwarded
DLP policy gate — configurable per tenant, never hard-coded
Model router — task-aware selection across Claude, GPT-class, Gemini, and Mistral
Zero-data-retention posture with every supporting provider
Response governance — output scanned before it reaches the employee
Immutable, hash-chained audit trail with automated tamper detection
TLS 1.3 in transit, AES-256 at rest, keys held in a managed KMS
Pricing · Early access

Priced for the governance,
not just the tokens.

Bring your own key and secured calls are unlimited. Use our Zero-Data-Retention inference — nothing is stored or used to train AI — and you spend credits, because the model cost is real. Start free, top up as you grow, and license Enterprise for the whole team. Launch pricing in INR, indicative during early access.

Free

Individual
₹0
forever · 1 user

A real taste of secured AI — plus an unmetered passthrough when you bring your own key.

  • 5 secured requests / day on our Zero-Data-Retention models (never stored or used to train AI)
  • Bring your own key — limited secured calls / month
  • Bring your own key — unlimited unmasked calls (gateway, audit & routing)
  • PII masking + prompt firewall on every secured call
  • 7-day personal activity log
  • India data residency · community support
Upgrade to Pro to buy credits & unlock unlimited secured bring-your-own-key
Most popular

Pro

Individual
₹1,499
/ user / month · incl. 1,000 credits

Fully compliant and secured — unlimited with your own key, metered credits on ours.

  • Everything in Free, plus —
  • Every call masked, audited & fully compliant
  • Bring your own key — unlimited secured calls
  • 1,000 managed Zero-Data-Retention credits / month included
  • All frontier models — router auto-picks the most cost-effective one
  • Premium models (Opus 4.8, Fable 5, Gemini 2.5 Pro, GPT-4.1) draw more credits · top up anytime, no cap
  • Secret scanning + DLP presets
  • Full audit trail + export · document & spreadsheet analysis
  • Email support
Coming soon: chat with your own documents (secure RAG), AI agents & secured connectors

Enterprise

Team
Custom
pooled credit wallet · volume pricing

Org-wide governance with a shared credit wallet and a signed Zero-Data-Retention guarantee.

Contact for enterprise license
  • Everything in Pro, org-wide, plus —
  • Pooled organisation credit wallet + volume credit pricing
  • Per-seat & per-team credit budgets in the admin console
  • Unlimited bring-your-own-key secured calls across the org
  • SSO / MFA + role-based access control
  • Row-level tenant isolation · custom DLP per tenant
  • Guaranteed Zero Data Retention — nothing stored or trained on — across every provider, with a signed DPA
  • Immutable, hash-chained audit ledger + tamper detection
  • India in-region / customer-VPC deployment
  • Dedicated onboarding + SLA-backed support
On the roadmap: chat with your documents (RAG), AI agents & automation, secured ERP / email connectors, private hosting
How usage works

Three ways to call a model — you only pay when we do.

Managed secured · zero data retention
Draws credits

Our inference, nothing retained

Runs through our full masking, firewall and audit layer to a frontier model under a Zero-Data-Retention agreement — nothing is stored or used to train AI. We bear the model cost, so it's metered in credits — 1,000 included monthly on Pro, a pooled wallet on Enterprise.

Bring your own key · secured
No credits

Your key, our guardrails

Your own provider key, wrapped in the same security and audit layer. No model cost to us, so it's unlimitedon Pro & Enterprise (limited on Free). Your provider contract governs retention.

Bring your own key · fast (unmasked)
Seat only

Skip masking, keep the gateway

Your key with no PII redaction — but still through our gateway: unified multi-model access, cost & usage logging, model fallback and spend caps. No token cost to us, so it's unlimited and covered by your seat — for prompts that carry nothing sensitive.

What a credit buys

Credits scale with the model — pay premium only when you need it.

Tier
Representative models
Rate
Economy
Drafting · extraction · classification
Claude Haiku 4.5 · Gemini 2.5 Flash · GPT-4o mini · Mistral Small · Llama 4
~1 credit / 1K tokens
Standard
Analysis · reconciliation · memos
Claude Sonnet 5 · GPT-4o · Gemini 2.5 Flash · Mistral Medium · DeepSeek-V3
~3 credits / 1K tokens
Premium
Deep reasoning · long context
Claude Opus 4.8 · Fable 5 · Gemini 2.5 Pro · GPT-4.1 / o-series · Mistral Large · Grok
~12 credits / 1K tokens

1 credit = ₹1. The router auto-selects the most cost-effective model that can do the job, so routine work stays on Economy / Standard and only genuinely hard tasks draw Premium. Managed Zero-Data-Retention inference only — bring your own key never spends credits. Top-up packs from ₹999 / 1,000 credits; volume discounts on Enterprise. Unused credits carry for 12 months.

Zero-Data-Retention posture acrossClaudeGPT-classGeminiMistral

Prices are indicative launch pricing for the early-access period and may change before general availability. GST applicable where relevant.

Frequently asked questions

Straight answers on data,
models, and compliance.

Do you train any model on our data?

No. FinIFRS does not train foundation models, and enables zero-data-retention with every underlying provider wherever it's offered. Your prompts and documents are processed to generate a response and are not retained by the provider for training.

Which frontier models do you support?

The router brokers requests to Claude, GPT-class models, Gemini, and Mistral today, chosen automatically by task, cost, and data sensitivity. The list is designed to grow — the governance layer sits above the model choice, not tied to any one provider.

Where is our data processed and stored?

Our default posture is India-first data residency for regulated finance customers — processing and storage pinned to an India region wherever the workload allows. Enterprise and regulated deployments can request dedicated in-region or customer-VPC configurations.

What happens if a prompt contains a customer's PAN, Aadhaar, or an API key?

The prompt firewall detects it before the request leaves the gateway. PII is masked according to policy; hard secrets such as cloud credentials block the request outright. Either way, the decision itself is written to the audit trail.

Begin

Give your team FinAI — frontier AI without the exposure.

Talk to our team about your data residency, DLP, and audit requirements — we'll walk you through the gateway architecture end to end.

Book Enterprise Demo →Back to Home