AI won't break your rules.
It'll walk right between them.

Cross-step correlation. Real-time action enforcement. Runs inside your app.

1. Run: npx skills add arcjet/skills
2. Add Arcjet protection to my app

Model: Totoro 1.302

1
const aj = arcjet({
2
  characteristics: ["userId"], // Track usage by user
3
  rules: [
4
    detectPromptInjection({ ... }), // Block prompt attacks
5
    detectBot({ ... }), // Block automated clients
6
    tokenBucket({ ... }), // Enforce budgets
7
  ],
8
});
9

10
export async function POST(req: Request) {
11
  const userId = "user-123"; // Get this from your session or auth
12
  const decision = await aj.protect(req, { userId });
13
  if (decision.isDenied()) {
14
    if (decision.reason.isBot()) {
15
      return new Response(
16
        "Automated clients are not permitted", {
17
          status: 403,
18
        });
19
    }
20
  }
21
  // Arcjet approved - continue processing
22
}

JSAstro Bun Deno Express Fastify Hono NestJS Next.js Node.js Nuxt React Router Remix SvelteKitPythonFastAPI Flask

Available SDKs.

Securing AI workflows at:

What your other controls miss

Unauthorized tool calls

The request looks clean. Then the workflow issues a refund, reads a file, or hits an internal API. Arcjet enforces at the action boundary, before it fires, not after.

Data exfiltration

Sensitive data slips out through prompts, tool outputs, and model context. Arcjet catches it inside the workflow, before it ever leaves the application.

Cost explosion

Runaway agent loops can burn a token budget in minutes. Without inline enforcement, the first you hear of it is the bill.

What Arcjet catches in real time

AI endpoint abuse protection

Cap tokens per user and org. Stop automated abuse before it reaches the model.

1
const aj = arcjet({
2
  // User/org scope
3
  characteristics: [
4
    "userId",
5
    "orgId"
6
  ],
7
  rules: [
8
    // Block scrapers and bots
9
    detectBot({ ... }),
10
    // Token budgets
11
    tokenBucket({ ... }),
12
  ],
13
});
14

15
// Block before inference
16
const decision = await aj
17
  .protect(req, {
18
    userId: user.id,
19
    orgId: user.orgId,
20
    requested: estimatedTokens,
21
  });

Prompt injection detection

Catch hostile prompts in user input and tool outputs before they reach the model.

1
const aj = arcjet({
2
  rules: [
3
    detectPromptInjection({
4
      mode: "LIVE"
5
    })
6
  ],
7
});
8

9
// Detect prompt
10
const decision = await aj
11
  .protect(req, {
12
    detectPromptInjectionMessage:
13
      userMessage,
14
  });
15
if (
16
  decision.isDenied()
17
  && decision.reason
18
    .isPromptInjection()
19
) {
20
  return new Response(
21
    "Forbidden",
22
    { status: 403 }
23
  );
24
}

AI data loss prevention

Strip sensitive data before it reaches model context, logs, or third-party tools.

1
const aj = arcjet({
2
  rules: [
3
    sensitiveInfo({
4
      mode: "LIVE",
5
      deny: [
6
        "EMAIL",
7
        "CREDIT_CARD_NUMBER",
8
      ],
9
    }),
10
  ],
11
});
12

13
// Catch before LLM
14
const decision = await aj
15
  .protect(req, {
16
    sensitiveInfoValue:
17
      userMessage,
18
  });
19
if (
20
  decision.isDenied()
21
  && decision.reason
22
    .isSensitiveInfo()
23
) {
24
  return new Response(
25
    "Sensitive data blocked",
26
    { status: 403 }
27
  );
28
}

Agent tool controls

Scope what every agent is allowed to do, by identity, role, and route, and enforce it in real time.

1
const aj = arcjet({
2
  characteristics: ["agentId"],
3
});
4

5
// Per-request
6
const ajForAdmin = aj.withRule(tokenBucket({
7
  rate: 1000, interval: "1m", capacity: 1000,
8
}));
9
const ajForMember = aj.withRule(tokenBucket({
10
  rate: 100, interval: "1m", capacity: 100,
11
}));
12
let decision;
13
if (role === "admin") {
14
  decision = ajForAdmin
15
    .protect(req, { agentId: agent.id });
16
} else {
17
  decision = ajForMember
18
    .protect(req, { agentId: agent.id });
19
}
20

21
if (decision.isDenied()) {
22
  return new Response("Unauthorized", { status: 403 });
23
}

Bring us your scariest AI workflow.Arcjet stops actions your stack would only log.

Start with code

1. Run: npx skills add arcjet/skills
2. Add Arcjet protection to my app

AI won't break your rules.
It'll walk right between them.

Embedded where the action happens

Enforced with runtime context

Security governance

Securing AI workflows at:

What your other controls miss

Unauthorized tool calls

Unauthorized tool calls

Data exfiltration

Data exfiltration

Cost explosion

Cost explosion

What Arcjet catches in real time

AI endpoint abuse protection

Prompt injection detection

AI data loss prevention

Agent tool controls

Bring us your scariest AI workflow.Arcjet stops actions your stack would only log.

AI won't break your rules.It'll walk right between them.

Embedded where the action happens

Enforced with runtime context

Security governance

Securing AI workflows at:

What your other controls miss

Unauthorized tool calls

Unauthorized tool calls

Data exfiltration

Data exfiltration

Cost explosion

Cost explosion

What Arcjet catches in real time

AI endpoint abuse protection

Prompt injection detection

AI data loss prevention

Agent tool controls

Bring us your scariest AI workflow.Arcjet stops actions your stack would only log.

AI won't break your rules.
It'll walk right between them.