Skip to main content

Getting started

PromptGuard sits between your app and LLM providers, scanning every request and response for prompt injection, data leaks, jailbreaks, and other threats before they reach your models. Add one line of code to secure every LLM call in your application:
import promptguard
promptguard.init()  # That's it — all LLM calls are now protected

Quick Start (5 min)

Add PromptGuard to an existing app in 5 minutes.

Tutorial (20 min)

Build a protected chatbot from scratch with step-by-step guidance.

Core features

Detection, redaction, and policy enforcement for AI applications. Add protection without rewriting your code.

Real-time Protection

Six-layer detection pipeline with ~150ms typical latency. Covers prompt injection, jailbreaks, PII, toxicity, and more.

Zero Setup Required

Works with any LLM provider — OpenAI, Anthropic, Google, Ollama, vLLM, Bedrock

Team & Compliance Features

Organizations, RBAC, SSO (OIDC), audit logs, and SOC 2 / GDPR compliance

Developer Friendly

Auto-instrumentation, Guard API, SDKs, CLI, and MCP server for AI editors

Security capabilities

14 specialized detectors across a six-layer detection architecture: normalization, regex patterns, ML ensemble, content safety classification, multi-turn intent drift analysis, and policy evaluation.

Prompt Injection & Jailbreak Defense

Multi-model ML ensemble plus LLM-powered jailbreak analysis across 7 attack categories

Content Safety & Multi-Turn Detection

LLM-based harmful intent classification and crescendo attack detection via semantic drift analysis

PII & Secret Detection

39+ PII entity types across 10+ countries with checksum validation, plus secret key detection with entropy analysis

Agentic & Output Safety

Tool injection detection for agentic workflows (OpenClaw, LangChain, CrewAI), streaming output guardrails, and MCP server security

Custom Policies & LLM Guard

Natural-language business rules, topic filtering, entity blocklists, and YAML policy-as-code

Real-time Monitoring

OWASP LLM Top 10 mapping, threat intelligence, token-level explainability, and compliance reports

How it works

Every request flows through PromptGuard’s six-layer security pipeline:

F1 = 0.887

Evaluated across 2,369 samples from 7 peer-reviewed datasets

99.1% Precision

Low false-positive rate validated in production workloads

< 200ms P95

Full ML detection pipeline latency overhead

Next steps

Quick Start

Add PromptGuard to an existing application in 5 minutes.

Detection Architecture

Read the whitepaper on the detection pipeline, benchmarks, and evaluation methodology.