About Marlow Systems

Marlow Systems is a vertically-integrated software house. We design, build, and operate production systems for ourselves and our clients across legal, beauty & POS, payroll, education, pharmacovigilance, fintech, and crypto. The work runs on hardware we own, with AI inference served locally and Bitcoin/Lightning available as a first-class payment rail.

Why owned hardware

• · Predictable cost. One-time hardware, no per-token billing, no surprise invoices when traffic spikes.
• · Data stays put. Customer data, model context, and inference logs never leave our network unless we explicitly push them out.
• · Faster iteration. Staging and production share the same hardware footprint — no “works on dev, breaks in prod” surprises from a hidden cloud capacity gap.
• · Hardened by default. Live containers run read-only, drop all capabilities, refuse new privileges, and bind only to loopback. Public traffic enters exclusively through Cloudflare Zero-Trust tunnels — no open ports.

How a project ships

1. Build locally against the shared blueprint — modern web stack, AI-aware from the first commit.
2. Push to staging on the LAN. A post-receive hook rebuilds the Docker stack and exposes the project for internal testing.
3. Promote to production on the same hook. The project becomes reachable on its public hostname through a dedicated tunnel, and starts pushing telemetry to our internal control plane.
4. The new project shows up on the operator dashboard alongside its peers, with backups, auth, observability, and AI-call attribution wired in on day one.

The shared bits

Some pieces are not per-project. They sit alongside everything else and every project uses them:

• · Inference proxy — every AI request from every container goes here first. Capability-aware routing, queue limits, per-project attribution headers so we know exactly which feature spent which tokens.
• · Local model serving — multiple model families served locally (text, code, voice, image, music, forecasting). Inference never leaves the building unless we explicitly opt-in to a cloud model for a specific task.
• · Bitcoin / Lightning rail — shared payment infrastructure that any project can plug into without spinning up its own node.
• · Encrypted backups + restore-test — every host backs up nightly to two destinations (local network-attached storage and an off-site cloud), with sha256 restore-test verification on every artifact before it counts as a successful run.
• · Resource monitor — continuous per-host telemetry and a smart kill-priority list so non-critical workloads cede memory to revenue-bearing ones under pressure.

This site is itself one of the projects we operate — same blueprint, same auth, same hardened deploy path as everything we ship for clients. Operator state and the live rack view live behind sign-in.