⚡ Trusted execution layer for AI & machine control
Execution, approvals, and automation control made safer.
Membara Commander connects AI, people, and real systems with bounded execution, clear approvals, and traceable audit trails.
Layered policiesAllow, approval, deny
Controlled integrationAI does not hold root authority
Traceable auditsDecisions and outcomes
Commander
System Online
Recent activitylive
Deploy configurationApproved
Restart servicePending
Read inspectionCompleted
Unsafe actionDenied
◇Execute safely.With control.
Core capabilities
Core controls for AI and automation.
From policy to evidence, each component helps keep real work bounded, approved, and auditable.
Policy Engine
Apply risk-based rules to decide which operations can proceed, need approval, or must be denied.
- Role & scope aware
- Risk classification
- Fail-closed defaults
Approval Flow
Hold sensitive operations before dispatch until the appropriate person approves.
- Exact-manifest approval
- Fresh human step-up
- Expiry & binding checks
Audit Trail
Record decisions, principals, effects, results, and evidence in a structured audit trail.
- Human/MCP separation
- Correlation trail
- Bounded evidence
Device Sessions
Connect runtimes and devices through authenticated, policy-bound sessions.
- Authenticated sessions
- Versioned protocol
- Deterministic rejection
Task Execution
Execute work with structured manifests, ownership fencing, and observable task lifecycles.
- Bounded execution
- Ownership fencing
- Recovery aware
Evidence & Logs
Capture execution evidence without exposing raw logs or secrets by default.
- Redacted outputs
- Checksummed evidence
- Clear recovery state
How it works
A clear path from intent to evidence.
Commander does not give AI unrestricted host access. Every task passes through identity, policy, a manifest, approval when required, bounded dispatch, and audited results.
Explore security principles →01
Request
An AI client submits a structured request.
02
Policy
Commander checks risk and scope.
03
Approval
Sensitive operations wait for human approval.
04
Execute
The Agent executes a bounded, fenced action.
05
Evidence
Outcomes, audit events, and recovery state are recorded.
Built around control
AI remains separate from human authority.
Human owners, MCP principals, device sessions, and the administrative recovery plane retain distinct identities and permissions.
✓
MCP cannot approveApprovals remain with the human owner.
✓
Root is not AI authorityAdministrative recovery remains out of band.
✓
UNKNOWN is not SUCCESSUnverified outcomes remain held for reconciliation.
Membara Commander
Start with clear controls, then scale deliberately.
The web app is being prepared for owner sign-in, task review, approvals, audit, and recovery.
Back to top ↑Google owner sign-in is planned after B05 runtime readiness.