CLAWSOLATOR ALPHABeta release planned · License under considerationRead what is verified
THE CLAWSOLATOR PRINCIPLEA prompt is not a permission system.Meet the independent daemonOperator authority belongs outside the agent.
Clawsolator home / Engineering guide / container isolation
Engineering guide / container isolation

Rootless Podman agent isolation: useful, measurable, and not magic.

Rootless containers help separate an agent process from the host's root account, but a container is not a permission policy. Secure agent operations require a composed design: identities, storage mounts, resource limits, networking, host mediation and tested execution paths.

Direct answer

Rootless Podman runs containers without requiring the container engine to execute as root on the host. When correctly configured, individual agent sessions or roles can be given distinct filesystems, resource budgets and network policies. This reduces some host risks, but does not eliminate shared-kernel threats, dangerous mounts, credential exposure or an overpowered management API.

What should be isolated for each agent?

FILESYSTEM

Private workspaces

Mount only the work a role needs. Avoid broad host-home mounts and shared secrets. Audit symlinks, ownership and persistent directories.

RUNTIME

Process and privilege boundaries

Run as an unprivileged host user. Keep privilege escalation, setuid paths, socket access and executable mounts under explicit policy.

RESOURCES

CPU, memory and processes

Set budgets and measure actual enforcement via cgroups and runtime inspection; do not assume an intended limit is active.

NETWORK

Outbound and inbound access

Default to the minimum required communications, constrain DNS and external destinations where supported, and treat exposed admin ports separately.

How XMAN differs from simply starting a container

A container can restrict execution; a management layer can also control who may create it, who can ask for more resources, who may change mounts, and when privileged host operations are reviewed. XMAN's described wrapper limits the operations exposed to agents rather than giving them unrestricted access to the Podman engine.

Why not grant the agent the Podman socket?

Access to a container engine's management socket can provide substantial control over workloads and host resources, even in rootless mode. The safer design is a narrow manager API with explicit allowed operations, path checks, separate approval for sensitive actions and audit records. The socket itself must remain protected and tested.

Does OpenClaw already support isolation?

Yes. Upstream OpenClaw documents Docker sandboxing and a rootless Podman deployment path. Clawsolator's proposed value is operator-facing coordination of governed roles, approval states, resource allocations and multiple gateways—not the invention of containers or a claim that OpenClaw provides no protections.

What should a meaningful isolation test prove?

  1. An agent cannot read another private workspace or the host home directory by default.
  2. An unapproved agent request cannot trigger a host script or acquire root authority.
  3. Assigned CPU, memory and process limits are actually enforced.
  4. Network destinations and service ports match the intended policy.
  5. Logs preserve the request, decision and verified result across a restart.

Failed tests should be documented with the affected configuration and fixed before the system is described as production-ready.

Independent technical answers / source-informed

Know what the agent can do. Know who decides.

Four concise paths through the architecture—written for operators, developers, security teams and evaluators. Each distinguishes configured controls from goals that still need testing.

NEW RESEARCH / DOCUMENTED OPERATOR FRUSTRATIONS

What users asked the systems to do

Firsthand issue reports, exact failure symptoms, the proposed engineering response and explicit verification limits.

Podman EACCES reports →Image lifecycle management →