Anteirus
Security and data for enterprise reinforcement learning

Train agents on your own work, without exposing it.

Anteirus is building the security and data layer for enterprise reinforcement learning. Environments are assembled from your own systems and your teams’ agent sessions, sensitive data is governed before a model sees it, and every episode leaves a record your security team can examine.

Talk to us See the architecture

In developmentAnteirus is founder-led and has not released a product yet. This page describes the approach and where it stands.

  • Built for
  • AI platform teams
  • Security teams
  • Owners of regulated data

Interface previewSample data. A design for the environment console, not a running product.

  • Data residency

    Designed to run where your data already lives.

  • Isolation

    Agents act in sandboxes, with networking and credentials you control.

  • Governance

    Sensitive fields are classified and masked before a model sees them.

  • Evidence

    Every episode, policy decision and reward is recorded for review.

These are design commitments for a product in development, not descriptions of a released service.

01 — Platform

Real work, made safe to learn from.

Agents improve fastest on tasks that look like your work, and those tasks live in systems full of sensitive data. Anteirus is designed to let you use the first without exposing the second.

Environments

Environments built from your own work

An environment pairs a real task with a sandbox to act in and checks that decide what counts as success. Anteirus assembles them from work your teams and their agents already do, so what an agent learns matches how you actually operate.

In development
  1. 01

    Tasks from real work

    Agent sessions, tickets and incidents become reproducible tasks with a known starting state.

  2. 02

    Sandboxed execution

    Agents act in isolated copies of your services, with networking and credentials you control.

  3. 03

    Reward checks you can read

    Tests, policies and human approvals decide success. Checks are versioned and reviewed like code.

  4. 04

    Repeatable episodes

    Each run is recorded step by step, so it can be replayed, compared and audited.

Data security

Sensitive data, governed

Before anything becomes an observation, a reward or a training example, it is classified, and policy decides whether it is used, masked or left out.

In development
  • Classification of sensitive fields
  • Masking before data reaches a model
  • Policy enforced at the point of use
Where the work happens

Anteirus Terminal

A terminal for working with coding agents. Each agent session is shown in full: prompts, tool calls, diffs, tests and approvals, with the context, changes and plan behind them.

In development · used internally
  • Each agent’s work readable in full
  • Context, changes and plan for every session
  • Many sessions side by side

Explore Anteirus Terminal

02 — Anteirus Terminal

Where agent work happens.

Anteirus Terminal is a macOS terminal for working with coding agents. Each agent’s session is shown in full, with a footer that says where it stands: its context, its changes and its plan. With hooks configured in the agent, it also shows cost and whether the agent needs you.

In developmentUsed internally at Anteirus since mid-2026. Not yet released, so there is no download.

Explore Anteirus Terminal Try the walkthrough

Interface previewSample data. One agent session and its footer.

03 — Architecture

From real work to evidence.

Anteirus is designed to run inside your boundary. Environments are built and run next to your data; at most, metadata leaves, and only if you allow it.

  1. Step 01

    Capture

    Agent sessions and selected tasks are recorded where the work happens, with what is needed to reproduce them.

  2. Step 02

    Classify

    Sensitive fields are found and labelled, and each dataset gets an accountable owner.

  3. Step 03

    Assemble

    Tasks, sandboxes and reward checks are packaged into environments your teams own and version.

  4. Step 04

    Prove

    Every episode, policy decision and reward is written to an append-only record you can export.

Trust boundary · self-hosted designCustomer data stays inside the boundary
Your environment
Repositories and services
Agent sessionsAnteirus Terminal
Data stores and warehouses

Read through policy

Anteirus data planeClassification · masking · environment build
Your RL environmentsTasks · sandboxes · reward checks · episodes
Metadata only, if you enable it
Anteirus control plane
  • Policy authoring
  • Environment registry
  • Audit export
Planned design The intended architecture of a product in development. In a fully self-hosted deployment, the control plane runs inside your boundary too.
Security

A named person responsible for security

Avnish Maha James Ranchigoda, Founder and Director, is responsible for Anteirus Technologies Corporation’s internal security work and vulnerability reporting. Contact: security@anteirus.com.

Security policy and scope

Engineering

Claims you can check

Every capability on this site carries its real status: in the current build, prototype, internal, planned, or a preview with sample data. The engineering notes describe how Anteirus Terminal handles your data.

Read the engineering notes

04 — Contact

Start with the work you want agents to learn.

Tell us which tasks matter, which data cannot leave your boundary and who must be able to review what happened. We will say plainly what Anteirus can do today and what it cannot.

Talk to us

How we handle what you send us is set out in the privacy notice.

Email
business@anteirus.com
Security reports
security@anteirus.com
Legal entity
Anteirus Technologies Corporation
Registered
Texas, United States. The founder works from Australia.