Arcjet
A runtime security platform for AI agents and web apps that runs inside your application code, blocking prompt injection, PII leaks, bots, and abuse at every tool call and request.
About Arcjet
Arcjet is a runtime security product integrated into application code as an SDK, covering AI agents as well as web apps and APIs. For AI agents, it can block prompt injection, prevent PII leaks, and make an ALLOW or DENY decision before every tool call; it supports hooks already provided by coding agents such as Claude Code, Copilot, Codex, and Cursor, or can be used via the SDK in self-built agents, and it records the policy behind each decision along with the values it read, making it easy to reconstruct what happened afterwards. For websites and APIs, it provides bot protection, Shield WAF, email validation, and rate limiting; rate limit rules are configured directly in application code, with state maintained by the Arcjet Cloud API, so there is no need to deploy extra infrastructure such as Redis. According to official figures, local decisions take under 1ms, and about 20 to 30ms when the cloud API is needed; the product has completed a SOC 2 Type II audit (unqualified opinion, covering security, availability, and confidentiality). The SDK is designed to fail open, so service outages or misconfiguration will not block all requests, and the SDK also fails open after its default 2000ms timeout.
Best for
Best use cases
Engineering teams that need to add ALLOW/DENY protection and audit records to tool calls in self-built AI agents.
Developers using coding agents such as Claude Code, Copilot, Codex, and Cursor who want to add security decisions on top of their hooks.
Teams that want to configure rate limiting, bot protection, WAF, and email validation inside application code to protect websites, web apps, and APIs.
Organizations that need to find out which agents are running in their organization with zero code changes (the Observe feature).
Compare before choosing
According to the official comparison page: Cloudflare's WAF, bot management, rate limiting, and AI Gateway evaluate inbound traffic at the edge, while Arcjet runs inside application request handlers and in agent tool calls the edge cannot see — the two are more complementary than replacements for each other.
According to the official FAQ: an AI control plane or MCP gateway is adopted at the organization level and sits outside the application, so at decision time it cannot see the parameters about to be passed to a tool; Arcjet is a package embedded in the code that performs the action, so at decision time it has the user, the input, and the run history.
According to the official comparison page: AI security posture management is used to discover assets and assess their exposure, whereas runtime security evaluates live activity and can intercept actions at the integrated enforcement boundary.
Features
Agent tool call protection
Delivers an ALLOW or DENY decision before every tool call — usable via the SDK in self-built agents, or through the existing hooks of Claude Code, Copilot, Codex, and Cursor.
Prompt injection & PII protection
Detects prompt injection, blocks PII leaks, and lets you add protection rules for tool, API, and database calls.
Audit records
Records what your agents actually did; every decision keeps the policy that ran and the values it read, so an answer can be reconstructed from it.
Agent discovery (Observe)
Discovers agents running in your organization without changing code, by sending OpenTelemetry data or connecting the Claude Compliance API.
Application-level rate limiting
Configure rate limit rules in application code by path, client characteristics, and algorithm options; clients are identified by IP by default, or by other characteristics such as user ID. State is maintained by the Arcjet Cloud API and takes effect across multiple application instances.
Web & API security
Bot protection, Shield WAF, email validation, and more for websites, web apps, and APIs.
Pros
- Decisions happen inside the code that performs the action, so it can see the user, the input, and the run history, and make an allow or deny decision before every tool call.
- Covers both AI agent security and web app/API security (bot protection, WAF, email validation, rate limiting).
- Every decision keeps the policy that ran and the values it read, enabling post-hoc reconstruction and auditing.
- Official figures show local decision overhead under 1ms, and about 20 to 30ms when the cloud API is needed.
- Has completed a SOC 2 Type II audit (unqualified opinion, covering security, availability, and confidentiality).
- Supports 20 SDKs covering Claude Code, Codex, Copilot, and languages and runtimes such as JavaScript, Python, and Go — LLM- and framework-agnostic.
- Rate limiting requires no self-hosted infrastructure such as Redis; state is maintained by the Arcjet Cloud API and takes effect across instances.
- Fails open on service outages or misconfiguration, so it never blocks all requests.
Pricing Plans
Individual
- 1 team member
- 1 hour log retention
- Email support
Startup
- 2 team members
- 24 hour log retention
- Email + Slack support
* Pricing information is for reference only. Please refer to the latest announcement on Arcjet's official website.