Not RMM with AI features. An autonomous managed machine.

39 AI agents run inside an OODA loop, gated by ChangeSets, recorded in a tamper-evident evidence chain. Every alert triages itself. Every remediation simulates first. 93% of incidents close auto-heal — the 7% that escalate include the full reasoning chain.

Six phases. Every signal. No scripts.

Every telemetry event, customer ticket, security alert, and capacity signal enters the same Observe → Orient → Decide → Simulate → Execute → Learn cycle. The brain reasons over context; the policy gate evaluates risk; the executor applies a ChangeSet and signs the evidence record.

  • 39 AI agents
  • 31 sentinels
  • 5 BIS subsystems
  • 6 ChangeSet templates
  • 56 CIS safeguards
  • 93% auto-heal rate
  • <30s detect → contain
  • ∞ evidence chain

39 agents. Three tiers. Every one MCP-exposed.

Core agents drive the OODA loop. BIS agents specialize in business intelligence subsystems. Specialized agents handle domain-specific operations. Each registers capabilities through the platform contract — drive any of them from Claude Code, brainstorm CLI, or the in-product /console.

All · 39
Core · 20
BIS · 5
Specialized · 14

Orienter

Context assembly from signals + history

Decider

Proposes ChangeSets within policy guardrails

Executor

Applies ChangeSets, signs evidence records

PolicyGate

OPA evaluation, risk tiering, rate limits

EvidenceKeeper

Chain integrity + Merkle verification

Narrator

Human-readable prose for every decision

IncidentLead

Per-tenant incident orchestration

AlertCorrelator

Multi-signal de-duplication + grouping

RunbookCaller

Selects + invokes ChangeSet templates

TenantOps

Tenant lifecycle + entitlement management

FleetSync

Edge agent distribution + posture verification

AuditQuery

Cursor-paginated audit-chain reads

CapabilityRegistry

MCP tool catalog + version management

HealthAggregator

Sentinel-grid roll-up to /status

IdentityGuard

Operator-action attribution + scope

KeyEnvelope

KMS wrap/unwrap + key rotation

RetryQueue

Idempotent retry with exponential backoff

NotificationRouter

Operator paging via channel preferences

ChangeSetSimulator

Dry-run execution producing resource diffs

OperatorConsole

HAI-chat surface + God-mode tool exposure

IdentityExposure

Breached credentials + dark-web monitoring

AttackSurface

CIS benchmark execution + EASM mapping

RiskPropagation

Graph-based blast modeling on incident

InsurancePosture

Carrier questionnaire automation

BYODRisk

Per-endpoint exposure scoring

BackupIntegrity

Snapshot hash verification + restore drills

ComplianceEvidence

Per-framework evidence aggregation

CapacityPlanning

Workload forecast + resource trajectory

vCISOReporting

Quarterly executive posture briefs

PatchOrchestrator

CIS-aligned patch installation

EmailQuarantine

Inbound message classification + pull

HelpdeskTriage

Ticket classification + auto-resolve

RMMConnector

Multi-vendor RMM normalization

EDRAggregator

Cross-EDR alert correlation

LicenseAuditor

SaaS license usage + reclamation

SaaSDiscovery

Shadow-IT detection via OAuth + DNS

NetworkDrift

Per-tenant overlay configuration drift

TenancyBoundary

Continuous RLS boundary verification

ForecastAccuracy

Decision-quality regression on outcomes

No destructive op without simulation.

Every mutation flows through the ChangeSet engine. Intent → simulation → diff → policy control → execution → evidence. Rollback is tested before commit. Below: an example ISOLATE_ENDPOINT trace.

  1. Intent
    • Decider proposes one named ChangeSet template with args.
      ISOLATE_ENDPOINT { endpoint_id: "e_8a4f", reason: "lateral movement detected" }
  2. Simulation
    • Dry-run returns the exact resource diff.
      edr.endpoints[e_8a4f].state: active → quarantined
  3. Diff
    • Blast radius computed against tenant scope.
      scope: 1 endpoint · cross-tenant: false · reversible: yes
  4. Control
    • OPA policy gate. Risk tier evaluated. Rate limits checked.
      tier: high · autopilot: allow · rate: 3/10 in window
  5. Execute
    • Apply. Sign evidence. Chain into the audit trail.
      evidence.id: ev_a83f9b · prev: ev_a83f9a · sig: ed25519:...

Autonomy is tunable, not binary.

Per-tenant autonomy lives on three tiers. Same agents, same evidence chain — different gates on what auto-executes vs what asks the operator first. Move tenants up or down the spectrum as trust accrues; the audit chain captures every change.

Tier 01 - Supervised

  • Operator approves every action.
  • Auto-execute:
    • Read-only queries
    • Inventory probes
    • Sentinel health checks
  • Operator review:
    • Ticket creation
    • Endpoint isolation
    • Policy changes
    • Credential rotation

Tier 02 - Guarded

  • Low-risk auto. Anything risky escalates.
  • Auto-execute:
    • Read-only queries
    • Inventory probes
    • Ticket triage & enrichment
    • Connector token refresh
    • Patch installation (CIS-aligned)
  • Operator review:
    • Endpoint isolation
    • Mass account changes
    • Cross-tenant operations

Tier 03 - Autopilot

  • Policy-granted. Critical actions still gate.
  • Auto-execute:
    • Read-only queries
    • Inventory probes
    • Ticket triage & enrichment
    • Connector token refresh
    • Patch installation
    • Endpoint isolation (ChangeSet)
    • ROLLBACK_POLICY (ChangeSet)
  • Operator review:
    • Cross-tenant operations
    • Identity-system mass changes
    • Billing-surface mutations

Different architecture. Not different features.

  • Reasoning
    • Traditional RMM: Scripted runbooks. Fixed conditional logic.
    • Brainstorm MSP: OODA loop on every signal. Brain reasons over context.
  • Mutation safety
    • Traditional RMM: Direct execution. Rollback is a recovery procedure.
    • Brainstorm MSP: ChangeSet simulation first. Rollback is a precondition.
  • Audit trail
    • Traditional RMM: Logs exported at audit time. Best-effort retention.
    • Brainstorm MSP: Evidence chain is the data structure. Replayable from any cut point.
  • Tenancy
    • Traditional RMM: API-layer scoping. Cross-tenant bugs possible.
    • Brainstorm MSP: RLS at the database layer. Cross-tenant architecturally impossible.
  • Autonomy posture
    • Traditional RMM: On or off. One global setting.
    • Brainstorm MSP: Three tiers (supervised, guarded, autopilot). Per-tenant.

Authenticated. Driveable.

Every operator surface exposes the runtime, evidence chain, and ChangeSet log for that view. Sign in once; drive any tenant.