#13

ClawHost Cloud

Live

Open-source VPS provisioning platform with transparent pricing

From

$10/mo

MIT licensed; transparent Hetzner/DO VPS pricing

Security

Basic 7.5/100

Price Range

$10/mo – $350/mo

Free Tier

Integrations

0 platforms

Security Score: 7.5/100 — Basic

ClawHost Cloud is a very new platform (Feb 8, 2026) by single developer (bfzli). Core advantage: dedicated-VPS-per-user model provides strong tenant isolation. Fully open source (MIT, 126 GitHub stars). Excellent pricing transparency across 45+ plans. However, no security documentation, no SECURITY.md, no MFA, no backup guarantees. Cloud-init has concerning defaults (allowInsecureAuth:true, HTTP-only Nginx). OpenClaw installed from npm without pinned versions. Extremely early stage with single-developer sustainability risk. More a VPS provisioning tool than a managed security platform.

10 risk categories scored 1-10 × evidence weight. Based on our methodology, grounded in OWASP Agentic Security, NIST CSF 2.0, and CIS Controls.

Can anyone else see my data?4/10

Dedicated VPS per user provides strong tenant isolation by architecture. Homepage: '100% Owned Data: Your own server, your data. No shared infrastructure, no logs, no third parties.' Cloud-init shows per-instance gateway auth tokens. However, no encryption-at-rest documentation, no data retention specifics, and no clarity on whether the platform dashboard accesses user agent data.

Can someone take over my agent?1/10

No documentation on prompt injection defenses, sandboxing, memory integrity, or human-in-the-loop. Platform deploys vanilla OpenClaw on VPS. Cloud-init sets 'allowInsecureAuth: true' on the gateway — a red flag that could expose agent management to unauthorized access. No container isolation or code execution sandboxing.

Are my keys and passwords safe?2/10

Architecture means credentials live on user's own VPS (not stored centrally by ClawHost) — positive. But no secrets management, root passwords injected via cloud-init template, gateway tokens stored in plaintext JSON on disk. No credential rotation, leak detection, or encryption of credentials at rest documented.

Can my agent do things I didn't authorize?1/10

Zero guardrails, rate limiting, behavioral monitoring, or kill switch for individual agents. Instance-level start/stop/restart exist but no agent-level controls. Platform provides raw VPS with OpenClaw installed — no platform-level rogue agent prevention.

Can I lose my data or get locked out?3/10

Recently added 'Export your OpenClaw as a portable zip archive' (Feb 14 changelog). Platform fully open source (MIT) — no vendor lock-in. However, no backup policy, no SLA, Terms allow discontinuation 'with or without notice.' Launched Feb 8, 2026 by single developer — significant bus factor risk.

Will I get unexpected bills?5/10

Pricing transparent: 45+ server plans across Hetzner/DigitalOcean/Vultr with exact specs and monthly costs. '$10/mo Starting.' Terms: 'Services are billed on a fixed monthly basis.' 'Cancel anytime.' No hidden fees — users pay only server cost. However, no spending caps or usage alerts. 'Prices subject to change with reasonable notice.'

Who's responsible when something goes wrong?1/10

No SECURITY.md on GitHub, no security advisories, no incident response, no audit logging, no breach notification, no compliance certifications. Privacy Policy generic: 'appropriate technical and organizational measures.' Terms allow modification 'with or without notice.' No GDPR DPA. Single developer with no organizational accountability.

What if a tool or dependency gets compromised?2/10

GitHub reveals full stack: Firebase auth, Hetzner/DO/Vultr APIs, Cloudflare DNS, Polar.sh billing. Cloud-init installs OpenClaw from npm ('npm install -g openclaw@latest') without pinned versions or integrity checks. No SBOM, no dependency scanning. Latest CI build shows failure. Auto-updates claimed but mechanism undocumented.

Is the platform itself secure?3/10

Firebase magic link + Google/GitHub OAuth — reasonable but no MFA. Cloud-init sets 'allowInsecureAuth: true.' UFW configured (22/80/443), gateway binds to loopback only. But Nginx config only HTTP (port 80) — SSL via Let's Encrypt mentioned in README but absent from cloud-init template. No security testing, no rate limiting documented.

Can I trust what my agent tells me?0/10

No documentation on hallucination mitigation, output verification, approval workflows, or AI uncertainty. Platform deploys vanilla OpenClaw without additional trust layers.

V = VerifiedD = DocumentedC = ClaimedU = Unknown

Full root accessDedicated VPS per userOpen-source codebase (MIT)