server2agent

How it works

1. Install (5 minutes)

Code-signed MSI on each Windows Server. Outbound-443 only — no inbound ports, no firewall changes. mTLS pinned. Zero secrets stored on the host.

2. Alert fires

Prometheus, Datadog, or your existing Alertmanager forwards the alert. server2agent fingerprint-dedups and triages in 1–2 seconds.

3. Investigation (automatic, read-only)

Claude Sonnet investigates with catalog-signed read-only Playbooks: disk, services, event log, IIS, network. Every fact has a source.

4. Your approval

One card shows what, why, the risk, and the rollback plan. You approve — the action executes in a constrained PowerShell sandbox. Or don't — nothing happens.

5. Full audit trail

Every tool call, approval, and command is written to an append-only hash-chained ledger. Ready for SOC 2.

What you get

A complete AI-first operations layer for Windows servers. All plans include the agent, the signed playbook runtime, the audit trail, and the orchestrator. Higher tiers unlock approval flows, fleet operations, and compliance integrations.

Natural-language AI chat

Ask "which servers have disks over 90%" and get an answer in seconds. The AI runs read-only playbooks on your servers, groups findings by host, and summarizes in plain language.

Incident + human approval

AI proposes state-changing actions (service restart, config change). Medium-risk needs 1 operator approval, high-risk needs 2. You are always the last mile.

Fleet console

Run any read-only playbook across all servers — or a role (web, db, ad) — in parallel. Results come back grouped by hostname, one table, one decision.

Real Windows-first Agent

Native .NET 8 Windows Service. Auto-install via MSI, auto-update via signed R2 channel, auto-uninstall on admin delete. Works behind corporate proxies.

Signed Playbook catalog

Every playbook is HMAC-signed and enum-locked. Free-form PowerShell is structurally impossible. The AI cannot invent commands.

Audit log with hash chain

Every admin action and every playbook run is recorded with a SHA-256 chain. Tamper is instantly detectable. CSV export on Pro+.

Slack / PagerDuty / SMS

Critical incidents page Slack, PagerDuty, or SMS — per-organization routing by severity. Email always included.

Tenant isolation + SSO

Clerk organizations as the boundary; every query filters by customer_id. SAML/OIDC SSO on Scale for AD-integrated logins.

Remote uninstall

Admin clicks Delete → agent receives the signal on next heartbeat → MSI self-removes the service and certificate. No RDP session needed.

Designed for "what if the AI is wrong?"

Read-only investigations run automatically; disruptive actions never do.

Medium-risk actions need 1 human approval. High-risk actions need 2.

Playbook allow-list is signed and loaded fresh each boot. The AI cannot invent commands.

Kill switch disables all LLM calls in 1 second — your servers keep running untouched.

30-day PII purge on account termination + written destruction certificate.

AI that investigates your server alerts — and asks before it acts.