AI agents · early warning · open source

Agents that actually ship—
and ship safely.

AgentNeo builds AI agents and environmental monitoring and early-warning platforms and takes them all the way to production — and publishes the methods, the data and the safety layer behind them in the open.

The tooling behind these accounts is single-operator — runs locally, posts on behalf of one human.

Monitoring and early-warning platforms
Agents in production
Open source · open data

Agent Audit Gate MIT · OpenClaw plugin

The safety layer between an AI agent and the strangers it talks to.

Put an agent on a public channel — a WhatsApp line, a support chat — and anyone who messages it can prompt-inject it or pull out private data. You can't stop the model from drafting a reply; by then the draft exists.

So instead of sending it, we hold it — and a senior AI agent audits it first.

inbound→ held · audited→ released

Approval gates usually need a human. Here the reviewer is a senior AI agent — so it scales. Junior drafts, senior approves. Tested live against prompt-extraction, instruction-repeat and social-engineering attacks — none got through.

openclaw plugins install clawhub:agent-audit-gate

Open data & methods CC BY 4.0 · Zenodo · Hugging Face

The datasets and methods behind our satellite products are published with a DOI, so anyone can cite, reproduce and audit them.

The chassis

NEO Portable Model-agnostic · in development

The portable runtime our agents are built on — a single, vendor-neutral chassis you run yourself.

Most agent runtimes are a thin wrapper around one vendor's SDK. NEO Portable puts the model behind an interface — the same loop has run end-to-end on local and frontier models alike. Swap the brain, nothing else changes.

Every tool call passes through a held-by-default gate — nothing runs unless it's explicitly allowed. Safety designed in, not bolted on.

any model→ held-by-default gate→ runs locally · you own it

Cheap local models do the leaf work; a frontier model is lit up only when judgment is required — so it runs anywhere you do and costs nothing when idle. The safety membrane is the same pattern we open-source as Agent Audit Gate.