This is the file on Anuj — a one-man agent-ops unit running out of Bareilly, India, who taught a laptop to behave like a security team, a dev team, and a research desk at once. No cloud bill. No co-founder. Just a plan that kept escalating.
What the file actually says about how this got built — and why it didn't stop.
No investor deck. No hiring plan. Just a working theory: if a system needs a data center to be smart, it isn't smart — it's expensive. So the build started backwards, stripping every dependency that cost money until only the free tier remained.
Eleven providers got stitched into one failover chain so the inference bill stayed at zero while everything else scaled up around it. Every agent, every cache layer, every recon script runs on hardware that fits under a desk.
The security instinct came first, the automation came to feed it — audits turned into full assessment pipelines, and the pipelines turned into the whole M4ST stack.
Base of operationsBareilly, UP
Known methodJugaad-first, ship over plan
Primary disciplineAutonomous agent infra
Secondary disciplineOffensive security / OSINT
"built it because nobody would fund it"
Exhibit B
Evidence locker.
What's actually in the toolkit, catalogued.
ITEM 01
Agent Swarms
Multiple autonomous agents that split work and check each other's output — nobody's babysitting the loop.
ITEM 02
MCP Bridges
21 custom protocol servers connecting models to databases, browsers, and recon tools directly.
ITEM 03
Free-Tier Routing
An 11-provider failover chain that keeps inference running at zero rupees, quietly, all the time.
ITEM 04
Local Inference Rig
Quantized models running on consumer-grade hardware — no rented GPU in the loop.
ITEM 05
Semantic Cache
A vector-similarity cache that intercepts repeat questions before they ever hit an API.
ITEM 06
Recon & OSINT
CEH-track investigative tooling, used on real infrastructure with real authorization letters.
Exhibit C
Open cases.
Systems that are actually running, not just proposed.