Security-First Architecture. Not an Afterthought.
ClawPulse was designed by security architects from day one. Your agent infrastructure is your most sensitive asset — it runs your business, manages your data, and executes actions with real-world consequences. We treat it that way.
Security Architecture Overview
ClawPulse implements a layered defense model. No single point of failure. Every layer independently enforced.
Six independent security layers. All six must be bypassed to control your agents.
Threat Model
We identified the realistic attacks against AI agent infrastructure and designed ClawPulse to defeat each one.
Credential Theft
Attacker obtains a password or session token through phishing, breach, or brute force.
→Passkey-only auth eliminates passwords. Device-bound keys cannot be phished or stolen without physical access.
Session Hijacking
Attacker intercepts or steals an active session token to impersonate an authenticated user.
→Step-up auth blocks write operations even on valid sessions. Session tokens are short-lived and rotated frequently.
Command Injection
Attacker intercepts or replays a control command to execute unauthorized operations on your server.
→Ed25519 signatures verify every command before execution. Replay attacks blocked by timestamp inclusion in signature payload.
Data Exfiltration via ClawPulse
Breach of ClawPulse infrastructure exposes your API keys, credentials, or sensitive agent data.
→Zero secret storage — ClawPulse never holds your credentials. A breach of our infrastructure cannot expose your API keys.
Man-in-the-Middle
Attacker intercepts traffic between your server and the ClawPulse dashboard.
→Mutual TLS with certificate pinning. Both sides verify certificates. No trusted connection without valid mutual certs.
Insider Threat / Overprivilege
A team member with access takes unauthorized actions or accidentally breaks production.
→RBAC with least-privilege defaults. Step-up auth for destructive actions. Full audit log of every action with actor identity.
Public Internet Exposure
Your monitoring infrastructure is reachable from the public internet and becomes an attack surface.
→Tailscale/VPN native support. Deploy entirely behind private network — no open ports required.
Unauthorized Agent Control
An attacker gains access to your OpenClaw agents through a compromised monitoring interface.
→Layered defenses: passkeys + mTLS + signed commands + step-up auth. Six independent security mechanisms must all be bypassed.
Eight Security Features, Detailed
Every security feature exists because we identified a specific real-world attack vector. Here’s the technical detail.
Passkey Authentication
WebAuthn/FIDO2. No passwords to steal.
ClawPulse uses WebAuthn/FIDO2 passkeys as the default authentication mechanism. Your login credential is a cryptographic keypair bound to your device — the private key never leaves your hardware. There is no password to phish, no credential database to breach, and no way to authenticate from an unregistered device. Passkeys meet FIDO2 level 2 assurance and are supported by all major browsers and platforms.
- FIDO2/WebAuthn Level 2 assurance
- Device-bound private key — never transmitted over the wire
- Phishing-proof: credentials are origin-bound
- Touch ID, Face ID, Windows Hello, and hardware security key support
- TOTP fallback for environments without passkey support
Mutual TLS (mTLS)
Bidirectional certificate authentication.
Every connection between the ClawPulse dashboard and your OpenClaw server uses mutual TLS — both sides present and verify certificates. A rogue client cannot connect to your server, and your dashboard cannot be impersonated by a MITM. Certificates are automatically provisioned and rotated during the installation process. You never touch a certificate file manually.
- TLS 1.3 enforced — older versions rejected
- Both client and server present certificates
- Auto-provisioned certificates during install
- Certificate pinning on the dashboard client
- Automatic rotation before expiry
Ed25519 Signed Commands
Every action is cryptographically verified.
Every control command sent from ClawPulse to your OpenClaw server — restart an agent, modify a cron, change a config value — is signed with an Ed25519 keypair. Your server verifies the signature before executing anything. If the signature doesn't verify, the command is rejected. No replay attacks, no command injection, no tampering in transit. The mathematical guarantee: it is computationally infeasible to forge a valid signature.
- Ed25519 (Curve25519) — 128-bit security level
- Signatures include timestamp to prevent replay attacks
- Per-session signing keys with automatic rotation
- Server-side signature log for audit trail
- Rejection rate monitoring to detect attack attempts
Step-Up Authentication
Write operations require re-auth. Always.
Read operations (viewing dashboards, checking metrics, reading logs) require standard authentication. Write operations — restarting agents, modifying cron schedules, changing configuration, revoking access — require re-authentication at the point of action. Even if an attacker compromises your active session token, they cannot control your agents without completing re-authentication with your device-bound passkey.
- Separate read and write permission levels
- Re-authentication required for all state-changing operations
- Step-up challenges are passkey-only (no password downgrade)
- Challenge validity window: 60 seconds
- Step-up events logged to immutable audit trail
Zero Secret Storage
Your API keys stay on your server. Always.
ClawPulse never stores your API keys, provider credentials, or any sensitive configuration values on our infrastructure. Your OpenClaw gateway is the single point of secret custody — ClawPulse only reads metrics and session data. When ClawPulse makes a recommendation that involves an API call, the call is executed by your local agent using locally-stored credentials. We see aggregated results, never the keys that produced them.
- Zero credential ingestion — ClawPulse never asks for your API keys
- Metrics are aggregated on-device before transmission
- Telemetry payloads contain no credentials or raw content
- Your OpenClaw gateway is the sole secret store
- ClawPulse infrastructure is not in your blast radius if breached
Immutable Audit Log
Everything logged. Nothing deleted.
Every action taken through ClawPulse — authentication events, configuration changes, agent restarts, alert acknowledgments, API calls — is written to an append-only audit log with timestamp, authenticated actor identity, source IP, and result. Audit logs cannot be edited or deleted through the ClawPulse interface. They are exportable in standard SIEM-compatible formats (JSON, CSV) and can be streamed to external log aggregators.
- Append-only storage — no update or delete operations
- Logged fields: timestamp, actor, action, target, result, source IP
- Log integrity via hash chaining (each entry includes previous entry hash)
- Export in JSON and CSV for SIEM integration
- Webhook streaming to external log aggregators
Tailscale & VPN Native
No open ports required.
ClawPulse is designed to work entirely within private network boundaries. Deploy it behind Tailscale, WireGuard, OpenVPN, or any VPN overlay — your monitoring infrastructure never needs to expose a public port. The ClawPulse dashboard connects to your server through your VPN tunnel, and the server agent only listens on localhost or VPN-assigned addresses. Your OpenClaw server can remain completely dark to the public internet.
- Full functionality over Tailscale / WireGuard / OpenVPN
- Server agent configurable to bind to VPN interface only
- No inbound port required if using VPN tunneling
- Tailscale ACL integration for network-level access control
- Deployment guide for air-gapped environments
Role-Based Access Control
Least privilege by default. Always.
ClawPulse has three built-in roles: Viewer (read-only access to all dashboards and metrics), Operator (can acknowledge alerts and restart services), and Admin (full configuration access). Roles are enforced server-side — not just in the UI. A Viewer with a modified request cannot perform Operator actions. RBAC applies to both the dashboard and the API. Custom roles are available on Team plans.
- Three built-in roles: Viewer, Operator, Admin
- Server-side enforcement — not UI-only
- Per-endpoint permission matrix
- Role assignment logged to audit trail
- Custom roles available on Team plan
Deployment Models
ClawPulse supports three deployment models, each with different privacy and convenience tradeoffs. Security properties are maintained across all three.
Local Private
Dashboard and server agent on the same machine or local network. Most private option.
- ClawPulse server agent and dashboard on the same network
- No data leaves your infrastructure
- No external connections required
- mTLS still enforced for all local connections
- Ideal for single-server setups
VPN / Tailscale
Access your dashboard from anywhere through your private VPN. No public exposure.
- Server agent bound to Tailscale/VPN interface
- Dashboard connects through encrypted VPN tunnel
- Access from any device on your Tailnet
- Firewall blocks all public internet access to agent
- Recommended for remote monitoring scenarios
Cloud Hosted
ClawPulse-hosted dashboard with your agents reporting via secure authenticated tunnel.
- Managed dashboard at getclawpulse.com
- Server agent establishes outbound-only authenticated tunnel
- No inbound ports required on your server
- Zero secrets stored on ClawPulse infrastructure
- Best option for ease of use
Why ClawPulse Makes OpenClaw Harder to Hack
Visibility enables response.
You can only respond to threats you can see. ClawPulse gives you real-time visibility into who is accessing your system, what commands are being executed, and when unusual patterns appear. Without monitoring, attackers can persist in your environment for weeks undetected.
Immutable audit logs deter insider threats.
When every action is logged with actor identity and cannot be deleted, it creates accountability. Individuals with access are less likely to misuse it — and when misuse does occur, forensic investigation becomes straightforward.
Passkeys eliminate the most common attack vector.
The majority of account compromises start with credential theft — phishing, password reuse, credential stuffing. Passkeys eliminate this entire attack class. There is no password to steal. No credential database to breach. No phishing page that works.
Step-up auth contains breach impact.
Even if an attacker fully compromises a session token, step-up authentication blocks them from doing anything consequential. They can read dashboards but cannot restart agents, modify crons, or change configuration without completing a fresh passkey challenge on your registered device.
Zero secret storage protects your API keys.
Your Anthropic API key, OpenAI key, and other credentials are what attackers actually want. ClawPulse never touches them. If ClawPulse's infrastructure were ever breached, the attacker would find monitoring metrics — not the keys that power your agents.
“ClawPulse doesn’t just monitor your security — it IS security. Using ClawPulse actively makes your OpenClaw installation harder to hack.”
Standards & Compliance
ClawPulse aligns with industry security standards. Not because compliance requires it — because good security does.
FIDO2 / WebAuthn
Passkey authentication standard
CompliantTLS 1.3
Transport encryption standard
EnforcedEd25519
IETF RFC 8037 signature scheme
NativeOWASP Top 10
Web application security risks
MitigatedZero Trust Architecture
Never trust, always verify
Designed-inPrinciple of Least Privilege
RBAC with minimal default permissions
DefaultReady to run your agents with confidence?
Security is included in every ClawPulse plan — even the free tier. Start monitoring with the security architecture your agents deserve.