Arcjet

Product Introduction

1. Definition: Arcjet is a runtime security platform specifically designed for AI applications and agents. It operates as a server-side security layer that integrates directly into application code to enforce policies in real-time before an action is executed.
2. Core Value Proposition: Arcjet exists to secure the AI application runtime by providing developers with programmable building blocks to prevent abuse, data leaks, and unauthorized access. Its primary value is enabling real-time, in-line security for AI features like chatbots, agents, and APIs without requiring complex infrastructure changes.

Main Features

1. Runtime Security Enforcement: Arcjet's core functionality is its protect() and guard() functions, which are called within application code to evaluate requests and agent actions against a configured security policy. It works by intercepting requests at the framework level (e.g., in a Next.js API route or Express middleware) or agent tool calls, analyzing them against rules, and returning an allow/deny decision before the business logic proceeds.
2. Multi-Layer Threat Detection: The platform combines several detection engines into a single policy. This includes Shield for common web attacks (SQLi, XSS), Bot Detection using advanced client signals and behavioral analysis, Prompt Injection Detection using specialized AI models to identify malicious instructions, Sensitive Information Detection to redact or block PII/PCI data, and Token Bucket Rate Limiting for precise cost and usage control.
3. Unified Management via MCP (Model Context Protocol): Arcjet provides an MCP server that allows AI coding assistants like Claude Code and GitHub Copilot to directly interact with its security platform. Through this protocol, developers can use natural language to list sites, analyze traffic, create remote firewall-style rules, investigate threats, and get security briefings without leaving their IDE.
4. Framework-Native SDKs: Arcjet offers deeply integrated SDKs for over 15 major web frameworks and AI agent libraries including Next.js, Express, FastAPI, LangChain, and Vercel AI SDK. Each SDK provides a type-safe, idiomatic API (like arcjet/next or arcjet/python) that abstracts the complexity of security logic, making implementation a matter of adding a few lines of middleware or a function wrapper.

Problems Solved

1. Pain Point: Uncontrolled AI API Costs and Abuse. AI endpoints are vulnerable to automated bot traffic and prompt flooding, leading to massive, unexpected bills from providers like OpenAI or Anthropic.
2. Pain Point: Data Exfiltration via AI Features. AI models can inadvertently leak sensitive data from prompts (PII, source code, secrets) in their responses, violating compliance regulations like GDPR or HIPAA.
3. Target Audience: Full-Stack and Backend Developers building AI-powered features into web applications. DevOps and Platform Engineers responsible for application security and cost governance. AI/ML Engineers and Prompt Engineers developing and deploying agentic workflows and LLM applications.
4. Use Cases: Protecting a public AI chat API from being scraped by bots. Preventing a customer support chatbot from leaking credit card numbers shared in conversation. Authorizing and validating tool calls (like database writes or email sends) within an autonomous AI agent. Immediately blocking a DDoS attack or traffic from a malicious IP range targeting an application's login endpoint.

Unique Advantages

1. Differentiation: Unlike traditional WAFs (Web Application Firewalls) that operate as external network gateways, Arcjet runs inside the application runtime. This allows it to make security decisions based on full application context—like user IDs, session data, and the actual content of AI prompts—which is invisible to a network-level WAF. Compared to generic rate-limiting libraries, it provides a unified, AI-aware policy combining bot detection, content analysis, and cost control.
2. Key Innovation: The integration of security policy enforcement directly into the developer workflow via the MCP Server. This transforms security from an ops-centric dashboard activity to a developer-centric, conversational task. Developers can discover threats, deploy mitigations, and understand security posture through their AI coding assistant, significantly reducing the mean time to respond (MTTR) to incidents.

Frequently Asked Questions (FAQ)

1. How does Arcjet detect prompt injection attacks? Arcjet uses specialized machine learning models trained to identify patterns and techniques commonly used in prompt injection attacks, such as jailbreak attempts, role-playing prompts, and indirect injection. It analyzes the text content of prompts before they are sent to the LLM, blocking malicious inputs in real-time.
2. Can Arcjet be used to secure existing applications without code changes? Partially. For immediate protection against network-level attacks (e.g., blocking IPs, countries, VPNs), you can use Remote Rules via the MCP server or dashboard without code changes. However, for deep content analysis (prompt injection, PII detection) and per-user rate limiting, integrating the lightweight SDK is required.
3. What is the performance impact of adding Arcjet to an application? Arcjet is designed for low-latency, real-time decisioning. The protect() call typically adds single-digit milliseconds of overhead, as most rule evaluations are performed locally. Complex analyses like prompt injection detection are offloaded to Arcjet's high-performance cloud service, maintaining fast response times.
4. How does Arcjet's pricing work? Arcjet uses a usage-based pricing model tied to the number of security decisions (protect/guard calls) made. Pricing scales with application traffic, and details on tiers and decision costs are available on their public pricing page.
5. Does Arcjet replace a traditional Web Application Firewall (WAF)? Arcjet is complementary to a network WAF. It addresses a different layer of the stack—the application runtime and AI logic. For comprehensive security, it is recommended to use Arcjet in addition to a perimeter WAF, as Arcjet can enforce business-logic and AI-specific policies that a WAF cannot see.