PlantoOS Product

Medhara

The memory and governance core — persistent recall, capability enforcement, and deterministic lineage.

Why agents need memory and governance

  • Agents are stateless by default — every session starts from scratch, discarding all prior observations, decisions, and patterns.
  • Retrieval systems return nearest neighbors, not contextually relevant memory — leading to redundant or irrelevant recall.
  • There is no standard access control for agent memory — any agent can read any data in the shared context.
  • Debugging agent failures requires re-running the full chain with no guarantee of identical outputs.

What Medhara provides

Structured memory with provenance

Typed memory objects with versioning, TTL, scope, and provenance chains.

MedharaMemoryExplorer
892 objects
Search memories by content, type, or provenance…
Observationobs-7a3f

User login frequency decreased by 34% after policy update

Scope: ProjectTTL: 24hv1Confidence: 0.883 min ago
Factfact-2b8e

Fraud detection accuracy improves 12% when combined with behavioral signals

Scope: GlobalTTL: v4Confidence: 0.9615 min ago
Decisiondec-4c1d

Route high-risk transactions through dual-agent verification pipeline

Scope: TeamTTL: 30dv2Confidence: 0.911 hr ago
Intuitionint-9e5g

Correlation detected between API latency spikes and false-positive rate increases

Scope: GlobalTTL: 90dv1Confidence: 0.722 hr ago
Observationobs-6f2a

Agent retry rate spikes during peak trading hours (14:00–16:00 UTC)

Scope: AgentTTL: 12hv1Confidence: 0.854 hr ago
Factfact-8d4c

PCI compliance requires masking of all card fields in agent memory

Scope: OrgTTL: v7Confidence: 0.996 hr ago

Provenance Chain

Raw observation captured
14:15 UTC · data-enricher
Pattern matched across 23 runs
14:22 UTC · pattern-detector
Crystallized to Fact
14:28 UTC · medhara-core
Validated by risk-scorer
14:30 UTC · risk-scorer

Source Events

Run #4,812 — pattern match
Run #4,799 — data correlation
Run #4,783 — initial observation

Linked Workflows

Fraud Detection Pipeline142 runs
Risk Assessment Flow89 runs
Compliance Audit34 runs

Crystallization pipeline

Raw observations are transformed into structured facts and distilled intuitions.

MedharaCrystallizationPipeline

Memory Crystallization Pipeline

Transforming raw signals into durable knowledge

Raw Observations
247objects
Structured Facts
89objects
Distilled Intuitions
23objects

Raw Observations

User login frequency decreased 34% post-policy update

obs-7a3f·0.88
monitor-agent · 14:15 UTC

API response time spiked 2.3× during batch processing window

obs-8b4g·0.82
perf-monitor · 14:12 UTC

Retry rate in fraud-detector correlates with peak trading hours

obs-9c5h·0.85
data-enricher · 14:08 UTC

Structured Facts

Fraud detection accuracy improves 12% with behavioral signals

fact-2b8e·0.96·v4·3 sources
Significancehigh
Policy: PASS

PCI compliance mandates masking of all card fields in memory

fact-3c9f·0.99·v7·5 sources
Significancecritical
Policy: PASS

Distilled Intuitions

Correlation between API latency spikes and false-positive rate

int-9e5g·0.72·2 linked facts
MaturityEmerging
Significance: Medium

Agent autonomy correlates inversely with error rate in regulated domains

int-0f6h·0.68·1 linked facts
MaturityForming
Significance: Medium

Capability-based governance

Access to tools and actions is governed through deny-by-default capability policies.

MedharaCapabilitiesfraud-detector
4 Approved 3 Denied

Agent Permissions

fraud-detector v2.4.1 · Capability-based access control

Allowed Tools

query_db
read-only
classify_intent
ml-inference
fetch_user_profile
read-only
validate_schema
utility

Denied Tools

send_alert
Requires dual-confirm gate
delete_record
Not in agent capability set
external_api_call
Blocked for regulated agents

Policy Evaluation Trace

1
Org Policy
allow-internal-toolsAllow
deny-delete-operationsDeny
require-audit-trailEnforce
2
Team Policy
pci-data-maskingEnforce
allow-db-readAllow
3
Project Policy
allow-ml-inferenceAllow
max-concurrent-callsLimit
4
Agent Policy
deny-external-apiDeny
allow-profile-readAllow
Final Result
Approved· 4 tools granted

Full decision lineage

Every decision links to its trigger, data sources, policies, and outcomes.

MedharaLineageGraph Explorer
7 nodes · 7 edges

Lineage DAG

Command
Agent
Tool
System
Outcome
Command
POST /detect-fraud
Agent
fraud-detector
Tool
query_db
Tool
classify_intent
System
memory-store
System
policy-engine
Outcome
Alert Dispatched

Selected Node

Agent
fraud-detector

Node Metadata

Data Sourcewebhook · api-gateway
Policy Versionv3.1
Timestamp14:32:01.042 UTC
Linked Memorymem-0a7f3c, mem-1b8e4d
Execution Time245ms
Token Usage1,284
Modelgpt-4o-mini
StatusCompleted

Connected Nodes

← cmd-1 (Command)
→ tool-1 (query_db)
→ tool-2 (classify_intent)
← sys-1 (memory-store)

Policy-bound agents

Capability enforcement ensures autonomous systems operate within explicit boundaries.

Governance Flow

Policy-bound agent execution ensures autonomous systems operate within boundaries.

Policy Layer
Agents
Capability Policies
Approved Tools
Enterprise Systems
Denied — policy block

Autonomous systems require explicit boundaries.

Where it sits in the stack

PlantoOS Architecture

The stack relationship between apps, agents, runtime, and systems.

Applications

Products and workflows

Agents

LLM-powered autonomous units

PlantoOS Runtime

Execution · orchestration · control

Capability Layer

Policy enforcement and tool access

Medhara Core

Memory · governance · lineage

Enterprise + Public Systems

Databases, APIs, infrastructure

A new compute layer for systems operated by agents.

Key workflows

How data flows through the system in typical usage patterns.

Workflow 1

Input

Agent observation stream

Core Process

Medhara writes typed, versioned memory objects with provenance

Output

Persistent structured memory available across sessions

Workflow 2

Input

Agent recall request

Core Process

Medhara ranks by relevance + recency + access patterns

Output

Contextually accurate retrieval, not just nearest-neighbor

Workflow 3

Input

Policy evaluation request

Core Process

Six-layer stack evaluated, deny-by-default, with audit event

Output

Governed action with full provenance trail

Measured outcomes

↓ 20–35%

Redundant retrieval reduced

↑ 18–28%

Contextual recall accuracy

↓ 30–45%

Debugging time with replay

Indicative ranges from internal benchmarks and early deployments; results vary by workload, model, and infrastructure.

How it integrates

SDK-first integration — governed from the first line of code.

1

Install SDK

pip install medhara or npm install @medhara/sdk

2

Initialize client

Connect to your Medhara instance with your project key

3

Write & recall

Use typed memory operations in your agent loop

example
from medhara import MedharaClient

client = MedharaClient(project="my-project")

# Write a typed memory object
client.memory.write(
    content="Q3 revenue forecast revised to $4.2M",
    memory_type="observation",
    scope="finance-team",
    ttl_days=90
)

# Recall with context
results = client.memory.recall(
    query="latest revenue projections",
    limit=5,
    rank_by=["relevance", "recency"]
)

# Check lineage
lineage = client.lineage.trace(action_id="act_abc123")