Reasoning and continuity
Medha
Keeps relevant context, retrieves trusted information, chooses suitable intelligence and coordinates approved tools across ongoing work.
Platform Architecture
See How Control Is Built Into the Platform
Medha handles reasoning and continuity, SastraPDF handles document work, and MedhaOS keeps identity, permissions, policies, approvals and audit around execution.
Platform roles
Three Parts, Clear Responsibilities
Each part of the stack has a distinct job so reasoning, document execution and enterprise control do not blur together.
Document work
SastraPDF
Handles document search, extraction, comparison, transformation, redaction, review, signing, routing and other document operations.
Enterprise control
MedhaOS
Keeps identity, permissions, policies, approvals, tenant boundaries, auditability and execution controls around AI-enabled work.
Controls around the work
Control Where It Matters
The architecture connects intelligence to accountable users, permissions, evidence and execution boundaries instead of treating AI output as an isolated event.
Protect Sensitive Data
Use customer-boundary patterns for classification, filtering, de-identification, rehydration and controlled access.
Know the User and Role
Apply identity, role, tenant and route checks before exposing context or allowing actions.
Apply Policy and Approval
Decide what may be suggested, retained, routed, executed or escalated to an accountable person.
Keep a Reviewable History
Preserve requests, AI suggestions, approvals, policy checks, system events and outcomes for review.
Connect Enterprise Systems Carefully
Control access to document systems, identity providers, workflow tools, enterprise applications and systems of record.
Choose the Deployment Boundary
Place cloud, private-cloud, hybrid or customer-controlled components according to data sensitivity and operational ownership.
Privacy and security
Protect Data Before Reasoning
Sensitive information can be reduced, separated or kept inside defined boundaries before selected AI processing takes place.
Data Protection
Classification, filtering, tokenisation and de-identification can reduce unnecessary exposure before AI or tools are used.
Tenant Boundaries
Identity, roles and tenant-aware data access help separate users, organisations and workflow context.
Safe Execution
Sensitive actions can require permission checks, policy checks, approval and a reviewable history before execution.
How work moves
From Context to Action
The platform is designed for work that continues across documents, users, systems and decisions, not merely one question and one answer.
01 / 06Scroll to explore
Bring the Work Into Context
Connect the relevant documents, users, systems, policies and task state.
Protect Sensitive Information
Filter, tokenise or de-identify data according to the deployment configuration.
Retrieve and Reason
Use relevant retrieval, specialised capabilities, reasoning paths and approved tools.
Check Policy and Permission
Apply identity, permissions, policies, approval requirements and execution boundaries.
Execute the Allowed Work
Use SastraPDF or connected systems only for operations permitted within the defined controls.
Keep the Evidence
Preserve actions, suggestions, approvals, policy checks, exceptions and outcomes for review.
Evidence
Know What Is Proven
Public claims should distinguish what is implemented, what depends on the customer environment and what remains an operating responsibility.
Implemented Capability
Persistent context, document operations, tenant-aware controls, PII filtering, semantic caching and governed workflow foundations are represented only where supported by evidence.
Environment-Specific Configuration
Models, integrations, identity providers, data boundaries and deployment topology must be verified for each environment.
Customer Responsibility
Operating procedures, access decisions, approvals, support and compliance responsibilities remain explicit.
Validation Through Use
Claims should be tied to demonstrated workflow behaviour and available evidence rather than broad lifecycle labels.
Start With One Boundary
Discuss the identity, data, integration, deployment, approval and audit requirements around one meaningful workflow.