Know the Task
Identify the user, tenant, workflow, permitted scope and current objective.
Medha keeps relevant context across ongoing tasks, brings in trusted enterprise information and uses suitable models and approved tools so teams do not have to start from zero each time.
Long-running work depends on documents, prior decisions, permissions and changing task state. AI becomes less useful when those relationships disappear between interactions.
Documents, prior actions and system records are repeatedly reconstructed instead of remaining connected to the work.
Different tasks are pushed through the same model even when retrieval, rules or specialised capabilities would fit better.
Useful reasoning can become risky when the system is allowed to act without clear permissions or approval boundaries.
Repeated prompts and duplicated context increase model usage without necessarily improving the result.
Medha sits around approved AI models and enterprise systems to preserve context, choose suitable intelligence, bring in trusted information and coordinate permitted work.
Maintain relevant task, tenant, user, document and workflow context across longer-running work within defined boundaries.
Route work across suitable language, document, embedding, classification, extraction and specialised models.
Use semantic search and retrieval to bring approved documents, records and organisational evidence into the task.
Select a model, deterministic rule, semantic cache, tool or human checkpoint according to the task and policy.
Reuse relevant prior computation and approved context where appropriate instead of processing the same information repeatedly.
Coordinate permitted APIs, document operations and workflow actions while MedhaOS retains policy and approval control.
Medha prepares useful work while identity, permission and approval remain explicit.
Identify the user, tenant, workflow, permitted scope and current objective.
Assemble approved documents, records, prior decisions and relevant task history.
Select appropriate models, rules, retrieval paths and reusable context for each step.
Separate the reasoning, supporting evidence and proposed action so they can be reviewed.
Check identity, policy, tool permission, approval requirements and audit obligations before execution.
Medha is intended for enterprise work where context, evidence and accountable action matter.
Compare policies, contracts, forms and supporting evidence while preserving task and document history.
Bring together approved records, identify the exception and propose an accountable next step.
Add governed reasoning and document intelligence to existing products, portals and internal systems.
Evaluate Medha against a defined workflow, trusted information, clear controls and an outcome that matters.
Test reasoning, context, retrieval and tool coordination against the intended workflow.
Confirm models, integrations, data boundaries and retention rules for the implementation.
Begin with one meaningful use case and expand only when the results support the next step.
See how governance, architecture and a guided scenario fit around Medha.
Identity, tenant isolation, policy, approvals, auditability and execution control.
See how reasoning, governance, data boundaries and enterprise integrations fit together.
Follow a bounded scenario from evidence and reasoning through proposed action and approval.
Concise answers to common evaluation questions, using the same governed product and capability definitions as the rest of this page.
Medha is Sastra's foundational enterprise reasoning platform, designed to coordinate reasoning, contextual continuity, retrieval, model orchestration and governed execution across enterprise workflows.
Medha is designed for continuing enterprise work where context, documents, decisions and system interactions may span more than one prompt or session. It coordinates retrieval, contextual continuity, reasoning and controlled execution rather than treating every interaction as an isolated exchange.
Medha provides the reasoning foundation. MedhaOS provides the governance and execution-control layer around enterprise AI, including identity, permissions, policies, approvals, tenant boundaries, auditability and controlled model or tool execution.
Medha is designed for multi-model orchestration. Model selection can vary according to the task, sensitivity, cost, quality, latency or policy requirements where those routing choices are implemented and configured.
No. Reasoning capability and execution authority are separate concerns. An action can be evaluated or prepared by Medha while permissions, policies, approvals and execution boundaries determine whether it may actually be performed.