Trust

Written for the people
who answer for it.

A firm's risk committee should be able to read how Akshāyvāt behaves before anyone uses it. This page is for them.

A fixed process
around a flexible tool.

The software decides what the model is shown, which tools it may use and how it answers. The model never sets its own limits.

The machine reads.
A lawyer decides.

Every piece of AI work is kept as a record: the documents and passages it drew on, the model and prompt version, the reviewer and the time. Nothing is fileable. A named lawyer approves it, or it goes no further.

Sources
Document and page
Made by
Model and prompt version
Signed off by
A named lawyer
Recorded in
The audit chain

What holds,
stated as facts.

T·01

Ethical walls

A partner can wall a colleague off a matter. It disappears for them from lists, search, the assistant and the docket. Enforced in the database, not only on the screen.

T·02

Documents stay in your estate

Where the firm uses SharePoint, documents stay there. Akshāyvāt keeps the index and the record beside them, not a second copy.

T·03

Mail is read, never sent

Each person's assistant reads only their own mailbox and proposes which matter a message belongs to. A lawyer confirms every link. It does not send email.

T·04

Human Eyes Only

Mark an email Human Eyes Only and the assistant does not see it until a person releases it.

T·05

Propose, then confirm

Most changes appear as a card first, and nothing changes until a lawyer confirms. A read-only switch withdraws every tool that can change anything.

T·06

Document text is material, not instruction

Text from documents, mail and web pages reaches the model marked as material to be read. A defence against prompt injection, not a promise of immunity.

T·07

Research is screened

A web search is screened for the matter's client and party names before it leaves, and the first search in a conversation is shown to you to confirm. A name screen, not a general confidentiality detector.

T·08

Tamper-evident record

Sign-ins, matter access, document actions, AI runs and approvals are written to a hash-chained record. A retroactive edit is detectable.

T·09

Single tenant, per firm

Each firm has its own installation, database and sign-in. No firm's data is shared with another.

T·10

The model behind an interface

Models sit behind one interface, so the firm can change provider without changing how it works. Which models its people may use is the firm's choice.

T·11

Your data does not train models

Matters, documents, mail and conversations are not used to train any AI model, by Akshāyvāt or by the model provider, whose business terms exclude it. What learns is the firm's own library and memory, which people can read and edit.

T·12

DPDP posture

The company is the processor. The firm remains the fiduciary.

No badge
we have not earned.

This page claims no certification. Akshāyvāt is in private beta. When an audit or a certification is complete, it will be listed here with its date.

Private beta · 26 November 2026

Come and see it.

Opening on Constitution Day to advocates in India, by invitation.