Trident: Cloud and AI Agent Security Platform

Trident gives security and engineering teams a single platform to observe, test, and protect AI agents in production while continuously mapping attack paths across cloud infrastructure. Whether you’re tracing every prompt your agent makes, running automated red-team attacks before a release, or scanning your cloud for over-permissioned IAM roles that an agent could exploit, Trident connects those signals into one security graph.

Quickstart

Instrument your first agent and start receiving traces in under 5 minutes.

Core Concepts

Learn how agents, findings, the firewall, and the security graph work together.

Agent Security

Trace agents, run red-team attacks, and enforce runtime firewall policies.

Cloud Security

Map cloud assets, discover attack paths, and generate remediation PRs.

What Trident Does

Trident covers two interconnected attack surfaces that legacy tools treat separately: the cloud infrastructure your AI runs on and the AI agents that run inside it.

Observe

Full-fidelity traces of every prompt, tool call, MCP invocation, and approval — with PII automatically redacted before leaving your process.

Test

200+ automated attack scenarios covering the OWASP Agentic Top-10. Run pre-deploy red-team campaigns from the dashboard or your CI pipeline.

Protect

Runtime firewall with two-stage prompt and output scanning. Confirmed findings automatically become ban rules enforced within 5 minutes.

Get Started

Create a project

Sign in to app.usetrident.dev and create a new project. Copy your Public Key and Secret Key from the project settings.

Instrument your agent

Install the SDK for your language and call trident.init() (TypeScript) or vouch_sdk.init() (Python) once at startup. Traces start flowing immediately.

Review findings

Open the Findings inbox to see security issues discovered during tracing and red-team runs, each scored by OWASP and AIVSS severity.

Enable the firewall

Switch your agent’s LLM base URL to the Trident gateway endpoint to route every call through the runtime firewall.

Explore the Docs

TypeScript SDK

Install @vouch-ai/sdk and instrument Node.js agents with one function call.

Python SDK

Install vouch-sdk to instrument Python agents, LangChain, CrewAI, and more.

REST API

Trigger red-team runs, query findings, and manage firewall rules programmatically.

CLI

Use tridentctl to scan your local environment and run evaluations from the command line.

⌘I