Deploy to a fleet
The six steps to deploy DefenseClaw enterprise to managed Windows, Linux and macOS endpoints, from choosing a profile and agents to your MDM recipe, config, ensure and a ringed rollout.
Deploy the standalone profile in six steps. You need an MDM, configuration-management tool or administrator shell that runs commands as SYSTEM on Windows and as root on Linux and macOS. Concepts defines profile, lifecycle and ensure. Not sure you need this? Read Is this for you?.
1. Pick the profile
Use standalone unless you run Cisco Secure Client and want DefenseClaw as one of
its modules (Secure Client). On Windows and
macOS the profile defaults to Secure Client, so set enterprise.profile: standalone
(Two profiles).
2. Pick the agents
Each agent you protect is a key under guardrail.connectors in the config. Nothing
is protected until you list one.
| Route | Agents |
|---|---|
| Machine policy: hooks in the vendor's administrator-owned policy files | Claude Code, Codex, Cursor, GitHub Copilot, OpenCode (with the managed plugin) |
| Per-user: the guardian writes and repairs the hook in each enrolled user's config | Devin, Antigravity, Hermes, Amp, Kiro |
OpenClaw and ZeptoClaw are not supported here. The verified combinations are in the enterprise route table; how to list agents and users is in Plan a rollout.
ACP (Agent Client Protocol) editors are a separate route, in preview, and not a
guardrail.connectors key. The central config needs acp.enabled: true, a
binding for each editor and agent pair, and an ACP profile that allows both
(ACP configuration). Then enroll each
user with /opt/defenseclaw/bin/defenseclaw-gateway enterprise acp enroll --user --client --agent --profile (/opt/cisco/defenseclaw/bin/defenseclaw-gateway on macOS; Managed computers).
Enrollment refuses a pair the config does not allow.
3. Pick your MDM
Every recipe follows the same MDM contract and uses the
scripts in packaging/mdm.
| Tool | What you push |
|---|---|
| Intune on Windows | A Win32 app that wraps the Setup and your config, plus a PowerShell 7 app |
| Intune on macOS, Linux | One root script: the kit wrapper |
| Jamf Pro, Iru (Kandji) | The .pkg and kit scripts |
| Workspace ONE UEM | A Win32 app or the .pkg, with scripts and sensors |
| Configuration Manager | An application with the Setup and your config |
| Linux configuration management | The .deb or .rpm through Ansible, Puppet, Chef, Salt, apt or dnf |
| Any other tool | The kit wrapper, run as SYSTEM or root (What you deliver) |
The release files: DefenseClawSetup-Enterprise-Standalone-x64.exe on Windows,
defenseclaw-enterprise-<version>-linux-<arch>.deb, .rpm or .tar.gz on Linux,
and defenseclaw-enterprise-<version>-darwin-arm64.pkg on macOS. Verify the
signed checksums.txt once and pin each file by its SHA-256.
4. Deliver the config and the AI Defense key
The config is one YAML file only administrators can change. It is the only policy
the host enforces (Policy changes on a managed host).
The lifecycle installs it at C:\ProgramData\Cisco\DefenseClaw\etc\config.yaml
(Windows), /etc/defenseclaw/config.yaml (Linux) or
/opt/cisco/defenseclaw/etc/config.yaml (macOS). Start from
A minimal config. A first install on Windows
needs a config; on Linux and macOS an install without one protects no agents. How
each wrapper takes it: Deliver the config.
The Cisco AI Defense key is optional. It never goes in the config, an MDM script or
a command line; store it with enterprise secret set
(Deliver the AI Defense key).
5. Run ensure and read the result
ensure is the one action your MDM runs. It installs, upgrades, repairs drift and
applies a changed config; when the host already matches it reports "noop": true,
so it is safe on every check-in. Exit codes: success 0 (Windows and Unix), failed
and rolled back 1603 or 1, busy 1618 or 75, invalid arguments 1639 or 2
(Exit codes). Check a host by hand:
sudo /opt/defenseclaw/bin/defenseclaw-gateway enterprise linux status --json{
"ok": true,
"action": "status",
"profile": "standalone",
"installed": true,
"readiness": { "gateway": true, "guardian": true, "enumerator": true, "sensor_helper": true },
"coverage_complete": true,
"security_complete": false,
"warnings": [
{ "code": "hook_process_fail_open", "message": "claudecode: a killed PreToolUse hook or PreToolUse timeout is non-blocking; ..." },
{ "code": "hook_process_fail_open", "message": "codex: a killed PreToolUse hook or PreToolUse timeout is non-blocking; ..." },
{ "code": "unprivileged_user_namespaces", "message": "standard users can create user namespaces (user.max_user_namespaces=...): ..." },
{ "code": "unprivileged_ptrace", "message": "standard users can trace their own processes (kernel.yama.ptrace_scope=0): ..." },
{ "code": "security_incomplete", "message": "security_complete is false because of: hook_process_fail_open, unprivileged_user_namespaces, unprivileged_ptrace; ..." }
],
"errors": []
}On macOS use /opt/cisco/defenseclaw/bin/defenseclaw-gateway enterprise macos status --json.
On Windows, from an elevated prompt:
& 'C:\Program Files\Cisco\DefenseClaw\bin\defenseclaw.exe' enterprise windows status --profile standalone --json.
Read coverage_complete and security_complete (Health fields).
With Codex, Claude Code, Copilot, Devin, Hermes, Kiro or Antigravity enabled, installed hooks can provide complete coverage
while security_complete remains false: the agent can continue after a stopped
or timed-out hook. hook_process_fail_open names that residual risk, once per
enabled connector, with the connector name first. unprivileged_user_namespaces and
unprivileged_ptrace appear while the kernel allows them, and security_incomplete
lists the warnings that keep security_complete false.
6. Pilot in observe mode, then expand in rings
Deploy a pilot ring with guardrail.mode: observe. It records every decision,
including what DefenseClaw would have blocked, without blocking on policy verdicts.
When the pilot is healthy and you have reviewed the decisions, set
guardrail.mode: action and deliver the config again. Then grow in rings (pilot,
early, broad, everyone). Plan a rollout
has ring sizes and exit criteria.
Next steps
Threat model
Trust zones, how each boundary is protected, and the risks that remain.
Operate
Check status and health, read logs and events, and verify machine policy.
Upgrade, repair and remove
Upgrade, roll back, change the config, repair and uninstall.
Troubleshooting
Fix a deployment by lifecycle error code, MDM wrapper code, hook refusal code or symptom.
Enterprise deployment
What DefenseClaw enterprise hardening is, which profile to choose, what runs on each endpoint, and what a standard user can and cannot change on Windows, Linux, and macOS.
Plan a rollout
Plan a DefenseClaw enterprise rollout. Choose agents and users, pilot in observe mode, remove per-user installs, and expand in rings.