Michael detects, contains, and explains attacks on your software, servers, and AI systems in under two seconds. It costs less than your hosting bill.
No server? Vercel, Render, Netlify, cPanel and shared hosting are covered by the free dashboard: one install command, lockfile-accurate findings, and virtual patches that deploy themselves.
Built for the teams that never had a SOC
The tools that could help cost $60 to $230 per endpoint a year, or $150k+, and are sold to companies that already have a SOC.
attack projects run by one operator with an open-source AI agent, breaching 100+ companies at about $50 per site.
of the web exposed by PHP2Shell in a single week, alongside a GitLab CVSS 10.0 and a LiteLLM credential stealer.
servers with no security team. Their owners are searching “am I affected” at midnight.
per endpoint per year, or $150k+ per year for a platform.
per server per month, free forever for your first one.
Guard is live in this build. Sword, Halo, and Pulse share the same evidence and decision layer.
Runtime, network, web, and identity coverage with autonomous response inside your policy guardrails.
Proves what is exploitable on your own estate, under signed authorisation.
Verifies artifacts, provenance, update channels, and agent egress.
New advisories become detections, and coverage time is published for anyone to check.
Most events never reach a large model, so deep protection costs about the same as a coffee a month.
Instant, free, and deterministic.
ML tuned to what normal looks like on each server.
Picks the response for about $0.0001 per decision.
Used only to explain the hard cases in plain language.
Typed decision models made per-event AI about 400× cheaper, and eBPF made deep host visibility fit in 150 MB.
A deliberately vulnerable portal on the live host. Michael probes it, records real findings, applies a data-only virtual patch, and blocks the exploit through a separate proxy. The app underneath stays unpatched.
SQL injection accepts the login. Command injection returns LAB_CMD_MARKER. Traversal reads the login source.
All four vectors blocked, including a 300 KB body-cap test. Every block is written to the audit chain.
Revoke the rule and the exploit works again. Re-probe and the rule is back. That is the whole lifecycle.
Incumbents are moving up-market. Google bought Wiz for $32B and left the people who need protection most behind.
cloud security in 2026, growing to $59B by 2031
small business security spend
managed security service providers
serviceable per year across 8M SMBs and 150k hosts
The agent and its local console are open source under Apache-2.0 and free forever for one server. The control plane stays ours.
Every action is reversible. Michael acts inside the guardrails you set, keeps the evidence, and gives you one-click rollback. In the live lab you can watch the rule come out and the exploit return.
Each decision ships with its evidence and lands in a hash-chained audit trail. Editing a row breaks verification at that entry, and the console shows you the head of the chain.
Authorised offence only: your own estate, a signed scope, and read-only proof payloads. The live lab refuses any hostname that is not loopback.
One server, forever, with the open-source agent. Paid plans add the fleet, the decision brain, the autonomy, and the offence.
Yes. The MSP plan is $12 per server wholesale for hosting providers and MSSPs.
Typed decision models cost about $0.0001 per decision. Blended inference is around $0.13 per server per month.
eBPF made host-level visibility fit in a footprint a $5 VPS can carry.
Incumbents are consolidating up-market, as Google buying Wiz for $32B shows, and the bottom is open.
For the people who need protection most and can afford it least. Be one of the first ten design partners.