Use the useful pieces.
Existing mobile clients, skills, messaging connectors and plugins can remain available through a compatible bridge.
MULTI-AGENT SYSTEMS / NOT AN OPENCLAW WRAPPERThe framework is replaceable.
The framework is replaceable.
The boundaries are not.
XERXES SI is building a modular ecosystem for agent execution, synthetic intelligence research, infrastructure operations and controlled development—not a product whose future depends on one agent harness.
ORGANIZATIONXERXES SI
Synthetic intelligence research and software engineering
INTELLIGENCE & DEVELOPMENTXERXES CORE + SIDE
Research, hypotheses, design, experiments and evaluation
OPERATIONS CONTROLXMAN + CLOUD MASTER
Agent coordination, resource admission, machine capacity and operational observability
EXECUTION BOUNDARIESCLAWSOLATOR
Isolation, guarded access, controlled approvals, enforceable policy
REPLACEABLE RUNTIMESOPENCLAW · HERMES · NATIVE
Useful plugins, models, messaging surfaces and custom harnesses
WHY XMAN MATTERS
Resource decisions should not burn another model turn.
The daemon can handle routine allocations using saved policy, resource budgets, priorities and actual enforcement readback—without asking a master model to debate every permitted increase.
See the policy example ↗WORKER REQUEST UNTRUSTED INTENT
XMAN DAEMON ADMISSION • FAIRNESS • ACCOUNTING
ALLOW / QUEUEWITHIN BUDGET
OPERATOR REVIEWSENSITIVE EXCEPTIONS
PODMAN READBACK ACTUAL RESULT
LEDGER + RETRY SAFETY + AUDIT EVIDENCE
FRAMEWORK INDEPENDENCE
No sacred dependencies.
If OpenClaw’s plugins, update cadence, security model or community value stop serving the mission, XERXES SI can replace that adapter—or build a new harness—while preserving the management plane.
Independent execution backends.
Native rootless Podman today; investigate OpenShell and other policy runtimes where they improve enforceable isolation.
Build for the larger horizon.
SIDE and the cognitive research program are intended to support increasingly capable systems under testable operator controls.
THE NEXT STEP
The architecture is the investment.
Harnesses come and go. Reproducible engineering, verified limits and independent governance remain.
RUNTIME ARCHITECTURE
Control authority outside the model.
Agent / harness
OpenClaw, Hermes or a future XERXES runtime proposes bounded work.
↓ REQUEST / NO PRIVILEGE GRANTED
XMAN daemon / manager API
Independent rules evaluate target, policy, identity, limits and current state.
↓ ALLOW / REQUIRE OPERATOR / DENY
Rootless Podman execution
Approved worker jobs use limited mounts, credentials, CPU/RAM budgets and documented logs.
THE GOVERNANCE DISTINCTION
Agents can ask. They cannot appoint themselves administrator.
Isolation limits blast radius. Resource policies bound consumption. An operator-approval path handles sensitive changes. Readback checks whether the runtime reflects the result.
Current evidence covers bounded non-root resource changes. A privileged root executor is a future gated component and should be evaluated separately before deployment.
The XMAN daemon explained →