Powered by Medhara

Medhara

The memory and governance core behind PlantoOS. Persistent memory, capability control, deterministic lineage — the kernel that makes agent systems governable.

Agents scale decisions. Medhara scales control.

  • Persistent memory across time — agents recall, not re-gather
  • Capability governance — who and what can access, act, and modify
  • Deterministic execution lineage — every path reproducible
  • Compliance-ready auditability — audit artifacts as a side-effect of execution

↓ 20–35%

Redundant retrieval / token spend

↓ 30–45%

Debugging time with replayable lineage

↑ 15–30%

Multi-step task success rate

Indicative ranges from internal benchmarks and early deployments; results vary by workflow.

Medhara Core = 4 primitives

The building blocks that make agent systems governable, persistent, and auditable.

Memory Objects

Structured, versioned, and linked. Every memory entry carries type, scope, TTL, and provenance metadata from creation through every read.

↓ 20–35% redundant retrieval per session

Policies & Access

Capability-based access control evaluated at every boundary. Six-layer policy stack from org to action level, deny-by-default.

100% action-level policy coverage

Lineage

Every decision is traceable. Full DAG from trigger to terminal state, queryable via API, exportable in standard audit formats.

↓ 40–60% faster root-cause analysis

Crystallization

Raw observations → structured facts → distilled intuitions. Medhara progressively compresses experience into actionable knowledge.

↑ 15–25% contextual reasoning accuracy

From activity to institutional memory

Medhara progressively transforms raw observations into structured, actionable knowledge.

Memory Crystallization

How Medhara transforms raw activity into institutional memory.

Raw Events

Logs, conversations, actions.

Observations

Signals and detected patterns.

Structured Knowledge

Entities, links, and relationships.

Operational Memory

Reusable context across workflows.

Institutional Intuition

Systems improve decisions over time.

Medhara transforms activity into structured memory.

Every action has lineage

Full traceability from trigger to terminal state — every decision reconstructable.

Decision Lineage

Every agent action is fully reconstructable.

Prompt / CommandT+0ms

User initiates action

Agent ReasoningT+120ms

Intent parsed, plan formed

Tool CallT+340ms

API invoked with parameters

System ActionT+510ms

External system mutated

OutcomeT+680ms

Result verified and recorded

Full lineage stored in Medhara

Every action should be reconstructable.

How Medhara powers PlantoOS

Medhara sits at the kernel level — between agents and enterprise systems — enforcing governance and persisting memory.

Applications

Your products and workflows

Agents

LLM-powered autonomous units

PlantoOS Runtime

OS shell + execution engine

Medhara Core

Memory + Governance kernel

Enterprise Systems

Databases, APIs, infrastructure

Provenance

Boundaries

Replay

Forgetting / Retention controls

What changes when you add Medhara

A side-by-side view of agent systems with and without governed memory and lineage.

Capability

Without Medhara

With Medhara

Memory

Stateless per session. Agents re-gather context every time.

Persistent, typed, versioned. Memory survives across sessions with provenance.

Tool Access

Ambient authority. Agents can call any available tool.

Capability-bounded. Every tool call evaluated against policy stack.

Auditability

Manual log correlation. Reconstructing decisions requires custom work.

Full lineage DAG. Every action traceable from trigger to output.

Debuggability

Black box. Reproduction requires re-running the full chain.

Deterministic replay from any checkpoint with identical outputs.

Compliance

Post-hoc artifact assembly. Evidence gathered after the audit request.

Continuous audit artifacts. Compliance data generated as a side-effect of execution.

Where we are headed

Medhara's roadmap — from memory foundation to ecosystem readiness.

Phase 1

Capture & Recall

Memory foundation: typed writes, provenance tracking, ranked retrieval, TTL-based lifecycle management.

Agents retain and recall structured context across sessions

Phase 2

Governance & Policies

Capability enforcement: six-layer policy stack, deny-by-default evaluation, RBAC-like access control at every boundary.

Every agent action evaluated against declarative policy rules

Phase 3

Runtime Determinism

Replay and lineage: deterministic execution paths, checkpoint-based rollback, full action DAGs for root-cause analysis.

Any agent run reproducible from any checkpoint

Phase 4

OS Marketplace Readiness

Extensible apps and connectors: third-party integrations, app marketplace, governed plugin boundaries.

Ecosystem of governed, composable agent capabilities

Frequently asked questions

No. Medhara is a governed memory layer, not a storage engine. It manages typed, versioned memory objects with provenance, access control, and lifecycle policies. Vectors may be used internally for retrieval ranking, but Medhara's value is in the governance, lineage, and crystallization layers above raw storage.

Yes. Medhara provides SDK wrappers for Python and TypeScript. You initialize a client, connect it to your agent loop, and use typed memory operations. The SDK handles provenance tracking, policy evaluation, and lineage capture transparently.

Our intended model is hybrid: customers can choose between Medhara-hosted storage or customer-controlled storage (your own infrastructure). In both cases, encryption at rest and in transit is standard, and access is governed by the same policy stack.

Observability tools watch what happened. Medhara governs what happens. It combines memory persistence, capability enforcement, and lineage tracking into a single kernel — not just logging and dashboards. PlantoOS is the OS experience layer on top.