Powerful enough to act. Safe enough to trust.
Ojas can see across your whole organisation, so it is built to share with discipline. Information moves by role and purpose, agents earn autonomy with evidence, and your people keep the final say.
Six rules every agent follows.
Need to know
An agent receives only what its current task requires.
Purpose-bound
Every share carries a purpose, and data can’t be reused for another one.
Least privilege
No agent ever has wider access than its human manager.
Summarise first
Totals, trends and bands are shared in preference to raw records.
Mask personal data
Phone numbers, emails and IDs are hidden unless the task needs them and consent allows.
Log every share
Who asked, who shared, what, why and when — visible to your compliance team.
Different data, different rules.
Public
Product information, published policies, price lists
Any agent may use
Internal
Department metrics, tasks, project status
Shared inside the organisation; summaries across departments
Confidential
Customer records, deals, pricing, conversations
Owning department in full; others see summaries or masked data
Restricted
Salaries, HR cases, bank details, board papers
Owning function only — never shared with other agents
Personal data
Names, phone numbers and emails of customers and employees
Consent-bound, masked by default, stored only in India
The sharing map, department by department.
This is the default. Every organisation can tighten it further.
Why does the Chief of Staff see only summaries? Leadership needs to know that Billing sentiment fell, not to read every customer’s conversation. Drilling into records goes through the department head, and is logged.
Agents earn the right to act alone.
Every agent starts by watching. It moves up one level only on a sustained record of accepted work with no compliance exceptions, and only when its manager approves.
- L0Observe
Watches and learns. Every new agent starts here.
- L1Advise
Suggests and drafts; people act.
- L2Act with approval
Prepares the action; a named person approves it.
- L3Act and notify
Acts within set limits, then tells its manager.
- L4Autonomous
Proven, high-volume, low-risk work only, checked by sampling.
Built in, not bolted on.
Consent captured per purpose. A withdrawal reaches every agent immediately.
All memory, search indexes and logs are stored in India.
Money and customer-impact limits for every agent, such as discounts and refunds.
Personal and payment data masked in transcripts, summaries and shares.
Every action, decision, share and refusal is recorded and reviewable.
Any department head can pause an agent instantly. The CEO can pause them all.
In-country models first. External models only as a fallback, never on restricted data.
App credentials in an encrypted vault; write access off until you switch it on.
When we work in your other AI tools.
Setting up and maintaining third-party AI tools, and placing dedicated people in your systems, carry the same discipline.
- We work under a data processing agreement as your processor
- We check each vendor’s data terms before we recommend or set up a tool, and tell you where your data will be stored
- Our people get only the access their task needs, approved by you, removed when the work ends
- Every team member signs an NDA and passes background checks
- Everything the Ojas platform stores stays in India
- If you leave, your data is returned and then deleted, with written confirmation
- Infrastructure we run is monitored 24×7, backed up, and tested for disaster recovery
People see Ojas through the same rules.
Company cockpit, every department summary, approvals above department limits, agent performance.
Everything in their department and its agents; summaries from others, as the sharing rules allow.
Their team’s records, coaching insights and agent outputs.
Their own records, tasks and performance insights, and answers from the help agents.
Every agent’s role card, the audit trail and the sharing log — with personal data masked by default.
Only what an agent chooses to send them, under consent.