Get everything you need from a modern CDN — from effortless scale and built-in security to programmable control and full visibility.
Visibility. Control. Governance. For every AI call.
Fastly AI Runtime Control (ARC)
Stop guessing your AI risk and spend. ARC is a unified control plane that sits between your applications and AI models, giving you centralized visibility, governance, and control over all your AI traffic. Track real-time spend, enforce budget and rate limits, automate multi-provider failover, and benefit from low-latency edge performance.
Benefits
Take control of all AI and agent traffic
Control internal AI usage and govern your AI-enabled applications. Fastly’s AI Runtime Control lets you safely manage multi-provider complexity with real-time visibility, cost and access controls, plus automated failover to guarantee continuous uptime.
Audit 100% of your AI activity
Satisfy legal, compliance, and security requirements. Verify how customer data is being handled and trace exactly who requested what.
Get unified multi-provider metering
ARC consolidates AI traffic into a single pane of glass with real-time, granular tracking of input and output token counts, request volumes, and calculated costs mapped to virtual keys and specific client sessions.
Quickly isolate high-spending hotspots
Use real-time dashboards and leaderboards to identify exactly which applications, teams, or developers are consuming the most tokens and dollars.
Turn provider outages into a non-event
Keep your customer-facing AI features running smoothly through rate limits or upstream provider outages. If a primary model goes down, ARC transparently retries and fails over to your backup models at the edge, ensuring a continuous user experience.
Enforce monthly budget limits
Transition from reactive spending to proactive financial guardrails by applying enforceable monthly dollar-budget limits directly to individual virtual keys.
Stop runaway agents from blowing your AI budget or quota
Rate limiting per virtual key protects your upstream provider quotas from being exhausted by runaway autonomous agent loops or external traffic spikes.
Protect against LLM-based threats with AI Firewall
AI Firewall is a security module within ARC that protects your AI-powered applications from prompt injection attacks with near-zero latency impact. It leverages Fastly’s SmartParse detection engine and custom guardrails to detect attacks like prompt injection and other LLM specific attacks. Deploy with a flip of a switch.
Explore AI Runtime Control
One control plane for all your AI traffic
Lock down access, cap spend, and stop attacks before they start. Fastly’s AI Runtime Control gives you one place to manage access, spend, and risk across every model and provider so you can run AI in production with real-time visibility and control.
Frequently Asked Questions
What is Fastly’s AI Runtime Control?
ARC is a control plane that governs AI traffic across your organization tracking real-time spend, enforcing access controls, and enabling automatic failover between AI models. It gives teams a single view into token usage, costs, and requests across every AI provider they use.
How can I control or reduce AI spend across my organization?
ARC tracks input/output token counts, request volumes, and calculated costs in real time, mapped to virtual keys and client sessions. Real-time dashboards and leaderboards show exactly which apps, teams, or developers are driving the highest spend, so you can spot and fix hotspots quickly.
What happens if my AI provider (like OpenAI or Anthropic) has an outage?
ARC keeps AI features and workloads running by automatically retrying requests and transparently failing over to backup models at the edge when a primary provider goes down.
How do I prevent a runaway agent from blowing through my budget or quota?
ARC applies rate limiting per virtual key, which caps how much any single application, agent, or user can consume. This protects your budget and upstream provider quotas from being exhausted by runaway autonomous agent loops or unexpected traffic spikes.
What is Fastly’s AI Firewall?
AI Firewall protects AI applications from attacks that specifically target models, like prompt injection attempts where a natural-language input tries to trigger unintended actions. Fastly's AI Firewall is purpose-built for these AI-specific threats, unlike a traditional web application firewall.
How does Fastly's AI Firewall detect and block prompt injection attacks?
It uses a three-step process: requests are first inspected by Fastly's SmartParse detection engine for LLM-specific threats, guardrails are injected to sanitize responses, and outputs are validated for refusal responses and leak protection.
What's the difference between AI Runtime Control and AI Firewall? Do I need both?
ARC is the control plane that manages AI traffic, cost, and reliability across providers. AI Firewall is a security layer that deploys inside ARC adding threat detection and guardrails on top. Most organizations run AI Firewall as part of their ARC deployment for both governance and protection.
Looking for more?
Fastly joins the Agentic AI Foundation
Fastly has joined the Agentic AI Foundation (AAIF) to ensure that the protocols governing agentic interactions are built for the scale and speed of real-world production environments
MCP at the Edge
Agent loops make dozens of tool calls per task. Learn how stateless MCP at the edge turns each one into a sub-millisecond hop.
Building an actually secure MCP server with Fastly Compute
MCP can be a security minefield if you build it wrong. This article provides a step-by-step walkthrough for deploying a secure, scalable MCP server on Fastly.