Arcjet

Arcjet is a developer-first runtime security platform for AI applications and agents. It is aimed at engineering teams shipping agents to production and security teams that need governance evidence. Its SDKs and Agent Runtime Security product discover running agents, apply policy to tool calls before they run, and keep an audit trail.

Arcjet is a San Francisco security startup founded in 2023 by CEO David Mytton. It began as an in-code security SDK for web apps, covering bot detection, rate limiting, the Shield WAF, email validation and PII detection, and has repositioned around AI agents. On 17 September 2026 it launched Agent Runtime Security, which has three parts. Observe inventories running agents by ingesting OpenTelemetry or polling Anthropic's Claude Compliance API. That needs no code changes, and it links an agent's actions across sessions into one workflow. Enforce uses Arcjet Guards, deterministic policies written in Rego on Open Policy Agent, which run before a tool call executes and can block it, require human approval, or explain the decision to the agent. Example policies block emails to external recipients or cap refund amounts. Audit keeps each action's inputs, policy decision and result as compliance evidence. Some controls, including a local prompt-injection model, run inside the customer's process, so sensitive data never leaves it. Native integrations cover the Claude Agent SDK, Claude Managed Agents, OpenAI Agents SDK, LangChain, Langfuse, Strands, Mastra and Microsoft Agent Framework. Hook-based policies also cover coding agents such as Claude Code and GitHub Copilot. The SDKs are at 1.0 for JavaScript/TypeScript, which supports 15+ frameworks including Next.js, plus Python and Go. The open-source arcjet-js monorepo is Apache-2.0 licensed. Arcjet says 500+ production applications use it. It has raised about $12M: a $3.6M a16z-led seed in 2024 and an $8.3M Series A led by Plural in October 2025. It competes with AI-guardrail and agent-security vendors on one side and with CDN/WAF bot management on the other.

Category

Governance & Security

Pricing

Freemium, Usage-based, Subscription

Target Market

Enterprise Developers, Security Engineers, CTOs, Platform Engineering Teams

Deployment

Cloud-first, API-based, Hybrid

Founded

2023

Headquarters

San Francisco, USA

Customers

500+ production applications (Sep 2026, per vendor)

• ✓
  Agent discovery (Observe)

  Inventories running agents from OpenTelemetry or the Claude Compliance API with no code changes, and links their actions into one workflow.

• ✓
  Arcjet Guards (Enforce)

  Deterministic Rego/OPA policies evaluated before a tool call runs, which can block it, require approval, or return an explanation to the agent.

• ✓
  Audit trail

  Keeps each agent action's inputs, policy evaluation and result, giving security and compliance teams evidence to reconstruct incidents.

• ✓
  Prompt injection detection

  Screens untrusted input for adversarial instructions before it reaches the model or a tool, including via a locally run detection model.

• ✓
  Sensitive data detection and redaction

  Detects card numbers, emails, phone numbers and other PII in process, so the data is blocked or redacted before it leaks.

• ✓
  Rate limiting and token budgets

  Token-bucket, fixed-window and sliding-window limits that can model per-user AI token budgets to stop cost-abuse attacks.

• ✓
  Bot protection and Shield WAF

  Separates good bots from scrapers and credential stuffers, and blocks SQL injection and XSS within the application runtime.

• ✓
  Remote rules and DRY_RUN mode

  Global rules for geo, VPN and IP ranges can be pushed without a redeploy, and every rule can be trialled in DRY_RUN first.

• •
  Authorising agent tool calls

  A support agent's refund tool is wrapped in a Guard that blocks refunds above a threshold and requires human approval before it runs.

• •
  Agent inventory for governance

  The security team points OpenTelemetry at Arcjet to list every production agent and its actions, with no engineering work, for an AI governance review.

• •
  Preventing PII leakage from agents

  In-process PII detection redacts customer card and phone numbers before an agent sends them to an external email or API tool.

• •
  Stopping LLM cost abuse

  Per-user token-bucket rate limits on a public chat endpoint stop scripted abuse from running up the model provider bill.

• •
  Securing a Next.js app in code

  A team adds bot detection, Shield WAF and signup email validation to route handlers without deploying a separate WAF or proxy.

List pricing is published. The free tier is a hard cap of 10,000 requests a month with no overage. Paid self-serve plans are $25/mo (Individual) or $299/mo (Startup), plus $5 per million requests. Agent-request pricing, SIEM export, SCIM, custom retention and single-tenant or private-VPC deployment are Enterprise-only and negotiated per contract.

This page was written from 7 sources, 5 on domains other than arcjet.com.

1. 1.arcjet.com — arcjet.comvendor
2. 2.arcjet.com — pricingvendor
3. 3.blog.arcjet.com — introducing agent runtime security observe enforce audit
4. 4.siliconangle.com — exclusive arcjet launches runtime security to track and cont
5. 5.helpnetsecurity.com — arcjet agent runtime security
6. 6.prnewswire.com — arcjet launches agent runtime security for production ai age
7. 7.github.com — arcjet js