The platform

LoQOS is where the guardrails live, below every agent.

An operating system for AI agents in finance. Limits are set once by people, enforced below the agent, and explained when they bind, so agents re-plan instead of guessing.

How it holds →

01Physics, not policy

Not policy an agent obeys, physics it cannot break.

Prompt-level guardrails are guidance a model can talk its way around. LoQOS moves them below the agent, built from proven, deliberately boring parts.

NixOS

Immutable image

The same system, byte for byte, every time it boots. Nothing drifts.

Firecracker

One micro-VM per agent

A misbehaving agent is contained to its own sandbox.

OPA

Policy kernel at the OS

Every tool call, write and spend passes one choke point no agent can bypass.

eBPF

Kernel-level egress

An agent can reach only what its mandate grants. Off-path has nowhere to go.


02Guardrails that guide

A guardrail should steer, not just stop.

Five promises LoQOS makes to the people who set the limits, and to the agents that work inside them.

  1. Limits bound outcomes, not every step.Agents are free in how they work, and held to where they may end up.
  2. Every denial says which rule applied, and how to re-plan.An agent that is stopped learns why, so it does not guess its way into the next wall.
  3. A safe, reduce-only path is always open.Whatever binds, there is always a way to make the book smaller and safer.
  4. Every guardrail is proven by trying to break it.A rule that has not been attacked is a hope, not a control.
  5. Its cost is measured, not guessed.You see what each guardrail costs in missed opportunity, and decide with that in view.

03Mandates

Humans approve the envelope once. Agents are free inside it.

The kill switch has no exceptions, and no agent can read or change its own limits.


04What runs on it

Our instruments, our own fund, and in time, your agents.


05When it ships

In three phases, each usable on its own.

  1. 01

    Consoles

    Mandates, limits and kill switches, set by people in one place.

  2. 02

    The OS image

    Every agent in its own sandbox, under the policy kernel.

  3. 03

    Outside agents

    Your own agents, held to the same limits as ours.

More detail, including the architecture, is shared with partners in conversation. See the roadmap


Put your agents under limits they cannot talk past.

Tell us what you are working on. A founder replies, and the first call settles scope. No prices, no packages; those come in the conversation.

A founder replies within one business day.