New Sakshi sandboxes are live. Try agent governance on synthetic data, in your browser. Get a sandboxarrow_forward
witness · Agent governance for India

Govern every AI agent,
and prove it.

The governance layer for AI agents in Indian enterprises. Register every agent, witness every decision on a tamper-evident chain, and produce the evidence your regulators ask for, in your own environment.

dns Runs in your VPC or on-prem public_off Data never leaves India verified Tamper-evident by design
For every agent you run InventoriedWitnessed BoundedProvable

The platform

One governance layer, across every agent and vendor

Not another model or dashboard, the accountable record beneath the agents you already run. Five modules, one tamper-evident spine.

REGISTER

Know Your Agent

A live inventory of every agent and model, owner, autonomy tier, provenance, verified identity. Nothing runs unregistered.

WITNESS

Flight recorder

Every decision hashed into a chain anyone can independently verify. PII is tokenized before it is ever stored.

BOUND

Bounded autonomy

Envelopes route each action to auto, human review, or block, with oversight telemetry and a drilled kill switch.

VIDHI

Regulator-mapped

Signed RBI MRM, DPDP, SEBI and IRDAI evidence packs, from live evidence, clause by clause.

VISHWAS

India-calibrated fairness

Declared-first bias screens gated on statistical significance, so they don't cry wolf.

Predict & prevent

The governed-agent lifecycle, not just a flight recorder

A record tells you what happened. Sakshi also certifies an agent before it ships, watches it in production, and catches the day it silently changes.

certify → watch → re-certify

01 · Certify

Fit-for-production certification

No agent ships without a signed, RBI-mapped certificate: red-team, fairness, evaluation gates, a formally-verified autonomy policy and verified identity, with an expiry.

02 · Watch

The Guardian, in production

A live monitor over the whole fleet that halts, escalates or steers the moment an agent drifts, is poisoned, or goes off-policy, every intervention on the record.

03 · Re-certify

On silent change

Sakshi fingerprints each agent's behaviour and flags the day it changes, and whether it was declared or silent, which sends it back through certification.

The loop nobody else closes for regulated Indian BFSI: certify before it ships, watch it in production, and a silent change sends it back through certification.

Where it applies

Any sector running AI agents. Deepest where it's regulated.

The same spine governs agents in any industry, our coverage goes deepest today in BFSI.

Banking & financial services

Banks, NBFCs, cards, lending, securities

Clause-mapped to RBI MRM, DPDP, SEBI and IRDAI, our deepest coverage, and the arc in the live demo.

Banking & finance arrow_forward

Healthcare & regulated data

Diagnostic, claims, patient-facing agents

DPDP-aligned by design: PII tokenized at ingest and never stored raw.

Healthcare arrow_forward

Government & public sector

Accountable, auditable automated decisions

RTI-ready evidence exports a citizen or auditor can verify for themselves.

Government arrow_forward

Any enterprise running agents

Telecom, logistics, technology, and beyond

If agents shape decisions in your business, Sakshi gives them an accountable record.

Enterprise arrow_forward

Regulatory evidence

The mappings no global platform builds for India

Each instrument is encoded as data and scored against your live evidence, a readiness score you can act on, and signed packs you can submit.

RBI MRM

draft · comment filed

Model inventory, validation evidence, AI disclosure & human handoff, red-teaming, a drilled kill switch.

DPDP

Act 2023 + Rules 2025

Rule 13 SDF due-diligence: DPIA from live evidence, minimization & localization screens, signed audit bundle.

SEBI 16C

binding · Feb 2025

Sole responsibility for AI/ML outputs, including third-party models. Vendor model identity, verified.

IRDAI

anticipated · ~Sept 2026

FREE-AI read-across skeleton, ready to re-cut the day the working-group report lands.

Integration

Two ways in, by how much code you want to touch

Sakshi never asks you to rewrite an agent, and never sees a raw identifier, values are tokenized at ingest.

SDK, a few lines

Wrap the model and agent calls you already make. Register once, then witness decisions and enforce autonomy envelopes from inside your process.

# pip install sakshi-sdk
from sakshi import SakshiClient

agent = sakshi.register("loan-agent", owner="Priya")
with sakshi.witness(agent) as w:
  w.disclosure(channel="app")
  w.action(outcome="approved")

Gateway, zero code

Put an un-instrumentable or third-party agent behind the interception proxy, or add the Sakshi MCP server to any MCP host, no change to the agent.

# pip install sakshi-mcp
# front any MCP server, witnessed + enforced
sakshi-mcp-proxy --agent-id <id> \
  -- npx @modelcontextprotocol/server-filesystem

# or add Sakshi as an MCP server
tools: enforce_action · witness_decision

Works with the agents and models you already run

OpenAI Anthropic Google Gemini AWS Bedrock LangChain LangGraph Google ADK CrewAI AWS Strands AutoGen MCP

Full developer docs & API reference arrow_forward

Backed by open research from Rota Labs: the governance models are Apache-2.0 on Hugging Face.

Deployment & data residency

Runs where your data lives

A governance product you can't trust is a contradiction. Sakshi is single-tenant, installs in your environment, and we never see your evidence, we couldn't hand it over if compelled.

  • Your VPC, your cloud, or on-prem (default)
  • Rota-managed in an India region, for teams without a platform crew
  • One-command install, Helm / docker-compose, nightly evidence backups

See all deployment options arrow_forward

Get your own Sakshi sandbox

We'll spin up a private, synthetic-data sandbox and email you the link: register an agent, witness a decision, generate an evidence pack. No procurement, no real data, no risk.