> ## Documentation Index
> Fetch the complete documentation index at: https://docs.protectorplus.cloudsine.tech/llms.txt
> Use this file to discover all available pages before exploring further.

# Threat Vector Database

> CloudsineAI's proprietary corpus of LLM attack patterns, maintained continuously by the CloudsineAI research operation.

The **Threat Vector Database (TVDB)** is CloudsineAI's proprietary corpus of LLM attack patterns. It powers the Vector Filter layer of [ShieldPrompt](/architecture/shieldprompt).

## What the TVDB contains

* Curated prompt-injection patterns across direct and indirect injection classes.
* Known jailbreak templates and their semantic variants.

## How the TVDB is maintained

CloudsineAI's research operation maintains the TVDB on a continuous basis. Emerging attack patterns are surfaced from public-domain disclosures, customer telemetry (consented), and CloudsineAI's internal research programme, then validated against current Protector Plus detection layers before incorporation.

<Frame caption="TVDB feedback loop — prompts and adversarial patterns flow into the threat database; live users gain protection against the latest threats.">
  <img src="https://mintcdn.com/cloudsineai-5cd7c547/GGS4yz3OwEKS-tHW/images/diag-tvdb-loop.png?fit=max&auto=format&n=GGS4yz3OwEKS-tHW&q=85&s=1f249d9499081f19559fc2913fa624cf" alt="TVDB feedback loop" width="1124" height="666" data-path="images/diag-tvdb-loop.png" />
</Frame>

## Cadence

The TVDB updates continuously, with accelerated releases for high-impact disclosures.

## On-prem & air-gapped delivery

In air-gapped deployments, TVDB updates are delivered as **signed bundles** through the customer's out-of-band update channel. The management console verifies and imports the bundle into the local vector store and makes it available to the Vector Filter without service interruption.