About Arcjet

What Arcjet does

Agents no longer just answer questions – they move money, change records, and ship code. Arcjet gives security and engineering one control plane for that risk: Observe, Enforce, Audit. Policy runs where the action happens, with the context a network proxy never sees.

Where Arcjet runs

Native SDKs for JavaScript, TypeScript, Python, and Go protect services, queues, scheduled jobs, and workflow steps – not only HTTP handlers. Coding agents such as Claude Code, GitHub Copilot, Cursor, OpenAI Codex, and Muse Code call Arcjet from the hooks they already fire. OpenTelemetry and the Claude Compliance API extend visibility without rewriting agent config. On developer machines, an optional Arcjet endpoint agent for macOS and Windows surfaces shadow AI connections, ties them to people and devices, and can block unmanaged use. See how to secure AI coding agents.

Observe

Know which agents are in play and what they did. Arcjet discovers agents and endpoints across the organization and attributes activity to people and devices, so investigations start with who acted – not only which site. Every in-app action is captured with full run context: user, session, actor, tool, typed arguments, and prior steps. Session behavior analysis flags risky sequences over time, and bot and agent fingerprinting call out automation, including clients that spoof a platform.

Enforce

Decisions land before the consequential step. Arcjet screens prompts for injection, detects and redacts PII and secrets across prompts, tool arguments, and responses, and applies coding-agent controls over destructive commands, MCP calls, and skills – allow, deny, or hold for review. Custom agents get the same mid-workflow answer before money moves or records change. Destination threat analysis scores the APIs, MCP servers, and URLs an agent is about to contact. The same platform covers classic application protection: Shield WAF, bot management, rate limiting, email validation, and signup abuse. See how to stop an AI agent contacting malicious URLs.

Audit

Every decision keeps the context that made it – policy version, actor, inputs, and run history – so you can prove what happened. On the Enterprise plan, stream that evidence into Datadog, Splunk, SentinelOne, Panther, or Amazon S3 for detection, alerting, and long-term retention. Arcjet has completed a SOC 2 Type 2 examination covering Security, Availability, and Confidentiality; the report is available in the Trust Center.

One policy, every surface

One policy engine – OPA Rego – covers coding agents, custom agents, and apps. Ship rules in code when you want them versioned, testable, and dry-runnable, or publish remote policies from the Console, API, CLI, or MCP. Framework integrations include the Claude Agents SDK, Claude Managed Agents, OpenAI Agents SDK, LangChain, Strands, Mastra, and the other stacks teams already ship on.

What makes Arcjet different

In-code decisions, not only at the edge

Arcjet runs inside your handlers and agent loops, so security policies can be evaluated on the full execution context. See tool call arguments, file paths, user session data, and other runtime context that would be invisible to edge-only controls.

Protects actions the network never sees

Tool calls, queued jobs, and fan-out work sit past HTTP. Compared with gateway-style controls such as Cloudflare AI Gateway or Datadog AI Guard observability paths, Arcjet can refuse those actions in process before they execute. Correlation IDs tie decisions in a session for investigation.

Low-latency analysis

Arcjet's SDK bundles a WebAssembly module for local-first analysis. Arcjet's cloud is deployed to over 300 edge locations for low-latency policy evaluation right next to users.

One platform for agents and HTTP

Coding-agent hooks, custom guard() calls, and HTTP protect() rules share one policy and evidence model. Alternatives that only wrap model providers or only sit on inbound HTTP leave the other half of the surface uncovered. Arcjet is built so security and engineering share one control plane for both.

Who uses Arcjet

• Security and platform teams enforcing policy on coding agents (Claude Code, Cursor, Copilot, Codex, etc) across developer laptops and cloud agents.
• Engineering teams shipping custom AI agents, MCP servers, and agentic workflows who need granular guardrails, not just prompt filters.
• B2B SaaS product and platform teams protecting APIs, signup forms, and AI endpoints with identity-keyed rate limits and bot detection.
• Developers who want security primitives as code — testable locally, reviewable in PRs — rather than a network appliance they cannot run in development.

Customers include Checkr, Creem, and JD Group. Arcjet is used by more than 1,000 developers across more than 500 production applications.

The team behind Arcjet

Arcjet Labs, Inc. was founded by David Mytton, CEO. David previously founded Server Density (acquired), co-founded Console (the console.dev newsletter), and researches sustainable computing at the University of Oxford. Follow him on X.

Arcjet started from a simple observation: production security tooling still behaved like network appliances, while developers shipped on Next.js, serverless, and now agents. The company raised a seed round led by Andreessen Horowitz in 2023 and Series A in 2025, with Plural, Seedcamp, and 20+ angels participating. We are a remote-first team of about ten across the US, Canada, and Europe, with a hub in New York's Flatiron district.

Company socials: GitHub, LinkedIn, X, YouTube, Discord.

Key facts

Frequently asked questions