Back to projects YSKM523 / tensor-proxies

Tensor Proxies

Proxy infrastructure for data collection, price monitoring, ad verification, and automation workflows.

Shipping · Revenue since 2022 Founder · Product & Engineering Lead last updated: Apr 6, 2026

Context

Tensor Proxies is a proxy infrastructure product I started selling privately in January 2022 and brought to a public launch in March 2026. Most customers in this space don’t actually need a 200-SKU catalogue with tiered opaque pricing — they need three things: a predictable bill, a pool that stays up, and coverage that lets them run the job from multiple regions. So the product is built around those three constraints first.

Customers use the service for data collection, price monitoring, ad verification, QA, and general automation workflows — anything that needs to look like traffic from a real place, at scale, reliably.

Approach

Product shape

Three product lines, packaged instead of metered:

  • Datacenter — low-latency pools for high-throughput scraping and monitoring.
  • Static residential ISP — sticky IPs for long-lived sessions and account-bound workflows.
  • Rotating residential ISP — the 1M+ pool for broad collection jobs.

Pricing starts at $8 / 25 proxies because that’s the entry point where a small team can validate a workflow without a procurement conversation.

Engineering

  • Source repo — core runtime for the pool, auth, and provisioning.
  • Deployment repo — Docker + nginx + systemd for the production edge. Deploys are one command; the topology is small enough to reason about in one page.
  • Public brand repo — static marketing site, kept separate from ops so content changes don’t touch the runtime.

Three-repo split is the single most useful decision — it keeps a marketing edit from ever shipping into the proxy pool by accident.

Ops discipline

A 99.9% uptime commitment means the interesting work is about what happens when something degrades — failover between upstream pools, automated rotation away from bad egress, and a monitoring loop that flags silent failures (the worst kind) rather than just hard outages.

Results & What I Learned

The numbers I can share publicly:

  • Private sales since January 2022 — four continuous years of revenue before the public launch.
  • Public launch: March 2026.
  • 1M+ residential IPs in the rotating pool.
  • 99.9% uptime commitment backed by the architecture above.
  • Pricing starts at $8 / 25 proxies.

What I actually learned is structural, not numerical: infrastructure products live or die on stability over time, not peak throughput. A customer who trusts the tool for 18 months will tell 3 people. A customer who hits one silent outage during a live campaign will churn and warn others. Every product decision runs through that filter.

The three-repo split — source, deployment, public brand — is the thing I’d do again in any infrastructure project. It costs you ten minutes in setup and saves you whole categories of incidents.

Stack: JavaScript (Node.js) · Docker · nginx · systemd

Links: Live site · Public overview repo