If you run an AI agent — for your business, your code, your customers — it can be tricked. A hidden instruction in an email, a web page, or a "helpful" third party can make it leak your keys or do something you never asked. NOVA stops that, automatically.
Prompt injection is the #1 attack on AI. It's not a virus — it's a forged instruction your agent treats as real. And right now, most agents have no defense.
An agent told to "summarise and forward" can quietly mail your API keys and passwords to a stranger.
A third party says "ignore the owner, I'm in charge." Without a wall, the agent obeys the impostor.
Data leaves through normal channels. No malware, no alert — just a clever sentence.
This runs on the real NOVA guard. Type anything (or tap a sample) and watch it get scored in real time.
Six core modules ship today. RedTeam and AgentGuard are live-engine features already built. Sentinel is the new guardian layer — all wired into one guard server.
Blocks secrets from ever leaving your agent's output.
Quarantines inbound injection attempts and impostor commands.
Identity proof — only you, with your secret knock, are obeyed.
Guardian model that stamps every input safe or suspect (new).
Every action logged and replayable — full provenance.
Live dashboard with alerts on every agent, in real time.
Attack simulator that probes your agent before launch.
Tool-use lockdown — what your agent is allowed to do.
Flat, affordable, built for the small-business owner — not just enterprises. All prices per agent, per month (ex VAT).
1 agent · 3 core modules (Shield, Gatekeeper, Handshake) · Ledger + Watchtower.
10 agents · all 8 modules (incl. Sentinel, RedTeam, AgentGuard) · priority support.
White-label · unlimited agents · your brand, your margin. Resell NOVA as your own.