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

# Trident — AI Agent Security and Cloud Risk Platform

> Trident gives engineering and security teams observability, automated red-teaming, a runtime firewall, and a cloud security graph for production AI agents.

Trident is a cloud and AI agent security platform built for engineering and security teams that deploy AI agents in production. It gives you full-fidelity observability into every LLM call your agents make, automated adversarial testing before you ship, a runtime firewall that blocks malicious prompts and outputs, and a cloud security graph that connects agent behavior directly to your cloud attack surface — all on a single data plane.

## Three pillars of agent security

<CardGroup cols={2}>
  <Card title="Agent Observability" icon="binoculars">
    Capture every LLM call, tool invocation, MCP interaction, and approval as a structured trace. Replay, diff, and analyze agent behavior at any point in time.
  </Card>

  <Card title="Automated Red-Teaming" icon="crosshairs">
    Fire 200+ adversarial attack vectors — prompt injection, jailbreaks, tool-call hijack, indirect injection, encoding bypass, MCP exploitation — against your agents before and after deploy.
  </Card>

  <Card title="Runtime Firewall" icon="shield-halved">
    Scan every incoming prompt and outgoing response in real time. Operator-confirmed findings auto-mint ban rules that propagate to the firewall within 5 minutes.
  </Card>

  <Card title="Cloud Security Graph" icon="diagram-project">
    Map cloud assets, identities, and data stores into a security graph. The toxic-combo engine finds multi-hop attack paths that chain a compromised agent through privileged cloud roles to sensitive data.
  </Card>
</CardGroup>

## How Trident fits into your workflow

Trident follows a four-stage loop that keeps your agents secure from development through production:

1. **Instrument** — Add the TypeScript or Python SDK to your agent in one `init()` call. Every LLM call is automatically traced with zero changes to your model code.
2. **Test** — Run automated red-team campaigns against your agent using Trident's attacker library. Findings are scored by OWASP category and AIVSS severity so you know what to fix first.
3. **Protect** — Route agent traffic through the Trident firewall gateway. The firewall runs a two-stage scan: your project's tenant deny-bank first, then LLM Guard scanners. Blocked patterns feed back into the ban-rule bank automatically.
4. **Respond** — Triage findings in the dashboard, replay attacks to verify fixes, generate remediation PRs with Sentinel, and monitor ongoing agent behavior with no-code Signals.

## Supported frameworks and providers

Trident auto-instruments agents built on any of these platforms — no manual span creation required:

| AI Providers                                            | Agent Frameworks                   |
| ------------------------------------------------------- | ---------------------------------- |
| OpenAI, Anthropic, AWS Bedrock, Google VertexAI, Cohere | LangChain, LlamaIndex, CrewAI, MCP |

## Get started

<CardGroup cols={2}>
  <Card title="Quickstart" icon="bolt" href="/quickstart">
    Instrument your first agent and see traces in the dashboard in under 5 minutes.
  </Card>

  <Card title="Agent Security" icon="robot" href="/agents/overview">
    Learn how to register agents, run red-team campaigns, and interpret findings.
  </Card>

  <Card title="Cloud Security" icon="cloud" href="/cloud/overview">
    Connect your cloud accounts and explore the security graph and toxic-combo engine.
  </Card>

  <Card title="TypeScript SDK" icon="npm" href="/sdk/typescript/installation">
    Install `@vouch-ai/sdk` and explore the full SDK reference.
  </Card>
</CardGroup>