Frequently Asked Questions
The stuff people actually ask us before signing up.
~ Can Deployment Bro break my server?+
Deployment Bro is designed to avoid surprise changes, not to act with full autonomy. Most fixes run through diagnostics, playbooks, and scoped server tools. If a command is needed, mttrly_execute_command normally creates a pending action; a user-authorized bounded Investigation is the explicit per-command exception. Both paths are audited and neither is raw unrestricted SSH.
~ Is this just ChatGPT for servers?+
No. ChatGPT gives generic advice. Bro has real-time access to YOUR server — logs, metrics, processes, services. When it says "94% RAM, node.js using 3.2GB", that's your actual server right now, not a guess based on training data.
~ What if AI hallucinates?+
The AI does not authorize its own risky action or bypass. Bro uses AI to choose diagnostics, playbooks, or scoped MCP tools; runtime policy decides whether an operation is read-only or approval-required. A bounded Investigation must be separately authorized, applies only to mttrly_execute_command, and is limited by server, inactivity, hard cap, and action count. Execution remains recorded in the audit trail.
~ Can Bro edit config files?+
Bro can read configs and recommend a change. Where mttrly_write_file is enabled, writes are restricted to allowed paths and always create a pending approval action. This is not raw write access: path policy, confirmation, and audit keep critical files under your control.
~ What's the difference between Watchdog and Bro?+
Watchdog: you give commands (/restart, /logs), bot executes after your action. Reliable, predictable, free. Bro: you describe problems in plain English ("why is it slow?"), AI diagnoses and suggests fixes. Pick buttons or AI assistance based on how you work.
~ Do I need Kubernetes?+
No. mttrly is designed for ordinary Linux VPS and cloud hosts where the required outbound agent can run. Common paths include systemd, Docker Compose, PM2, and journalctl; Kubernetes is not required.
~ What servers does mttrly support?+
mttrly targets Linux VPS, dedicated hosts, and cloud instances that can run the outbound agent and its required system tools. The most common tested paths use systemd, Docker, Docker Compose, PM2, nginx, and journalctl. Unusual managed platforms or custom distributions should be checked during setup instead of assumed compatible.
~ How much does it cost?+
Free: 1 server, watchdog mode, basic commands. Deployment Bro: $39/mo for 3 servers with AI and deploy pipeline. Need more servers? $15/mo each. No credit card to start.
~ Is mttrly secure?+
The required agent initiates an outbound encrypted connection, so mttrly does not require a new inbound management port. Risky actions are policy- and approval-gated. Relevant operational data and action history may be processed or stored to provide diagnostics and audit; the Privacy Policy describes the current data boundary.
~ What happens if mttrly goes down?+
Your applications keep running because the hosted control plane is not their runtime. During a mttrly outage, hosted diagnostics and actions may be unavailable, but your existing server access remains available. We do not publish a contractual uptime SLA on this page.
~ How do I get started?+
Create an account with email, add a server, and run the token-specific installer shown in onboarding. The outbound agent is required for inside-server diagnostics. Connect Telegram later if you want mobile alerts and approvals.
~ Can my team use it?+
Deployment Crew ($99/mo) is the team plan: 9 servers today, with shared team access rolling out in stages. Actions are logged so you know who did what.
~ Which messengers does mttrly support?+
Telegram is the Day 1 channel. Slack and Discord are limited beta channels for selected customers. WhatsApp is coming soon and disabled during private beta.
AI Safety & Limitations
Understanding how Deployment Bro works, what it can and can’t do, and how we ensure safe operation
~ Can Deployment Bro break my server?+
Deployment Bro is designed to reduce surprise changes, not to make production risk disappear. Three controls matter: (1) Runtime policy separates read-only playbooks from approval-required playbooks. (2) Scoped action layer prefers diagnostics and prepared playbooks over a free terminal; one-off command execution is policy-gated and audited. (3) Bounded exceptions such as an explicitly authorized Investigation session are limited by server, time, and action count. You still review evidence and remain responsible for changes to production.
~ Is it safe to give AI access to my production server?+
Deployment Bro does not use raw SSH as the normal AI control path. The required outbound agent executes scoped diagnostics, playbooks, and enabled MCP actions on the server. AI helps interpret the request and evidence; runtime policy controls what can run immediately and what needs approval. Operational requests, approval decisions, and execution results are recorded for review.
~ What if Bro doesn't understand my question or makes a mistake?+
Bro can misunderstand a request or interpret evidence incorrectly. If a request cannot be mapped to an available capability, use the capability list and inspect the underlying evidence instead of treating the answer as authoritative. Read-only questions do not silently become change requests. Risky actions follow the active approval policy, and a bounded Investigation session must be authorized separately.
~ How does the "show by default, change by request" principle work?+
A read-only request such as "show logs" or "check memory" runs through read-only capabilities and reports evidence. A change is represented as a separate action under the runtime policy. Approval-required actions wait for a decision; previously authorized automation or a bounded Investigation session follows its own explicit limits.
~ What data does Bro send to AI providers (OpenAI/Anthropic)?+
AI-assisted diagnosis can send your request and the relevant server output selected for that workflow, such as status data or log excerpts, to the configured AI provider. Redaction exists for common secret patterns, but you should still avoid placing secrets in prompts or logs. BYOK uses the provider configuration attached to your account. Provider handling and mttrly data retention are separate questions; review the current Privacy Policy and your provider terms rather than assuming zero retention.
~ Can I audit what Bro did on my server?+
mttrly records an operational audit trail for approval decisions, playbook and command execution, and results. The exact fields and available look-back depend on the current runtime and plan configuration. Export, webhook, syslog, and SIEM integrations are not promised on this page. Do not use the marketing site as a compliance guarantee; verify the current product behavior for your account.
~ What happens if Bro's AI analysis is wrong?+
Treat the AI explanation as a hypothesis supported by the displayed evidence, not as certainty. You can reject a proposed action, run another diagnostic, or investigate through your existing tools. Prepared recipes make the evidence-gathering path more predictable, while runtime policy and approval gates limit what a wrong interpretation can execute.
Still have one practical question?
Connect one server on the current free tier, or start with the public evidence visible from outside.