At Vercel’s recent Ship conference, CEO Guillermo Rauch made it clear: “Each new generation of software needs a new generation of infrastructure. For the agent era, that’s Vercel.”
With agent-triggered deployments skyrocketing from under 3% to over 50% in just six months, Vercel just dropped a massive wave of new agentic infrastructure, including the Vercel AI SDK, Vercel Services, the open-source EVE agent framework, and Vercel Connect.
But as autonomous agents in your environment are granted the power to execute shell commands, read files, and connect directly to your enterprise data stacks via native connectors, a critical question emerges: How will you secure your autonomous AI agents in production?
While Vercel’s new infrastructure excels at developer ergonomics, sandboxing, and token routing, its AI Gateway (AI GW) is fundamentally built for model access and FinOps. It is not designed for comprehensive AI security. By design, it doesn’t support third-party guardrails or threat protection.
That’s where Noma comes in. Today, we are excited to announce Noma’s native integration with Vercel Services, bringing industry-leading AI Detection and Response (AIDR) to the Vercel Agent Stack.
The Vercel Agent Era is Here: What You Need to Know
Vercel’s June 17th announcements redefined full-stack agent deployment. The core of Vercel’s new AI stack includes:
- EVE Framework: An open-source framework built on top of the Vercel AI SDK. EVE isolates agent execution by providing each agent with its own secure sandbox for shell commands, scripts, and file operations. It also includes tool-level configuration for human-in-the-loop approvals to pause destructive actions.
- Vercel Connect: A dedicated credential broker for AI agents. It features Vercel Managed Connectors for native integration into Slack, GitHub, Snowflake, and Salesforce, as well as Customer Managed Connectors for custom OAuth/API keys, enabling seamless “On-Behalf-Of” user authentication.
- Advanced Telemetry: Every agent run outputs OpenTelemetry traces covering prompt/response logs, model reasoning steps, tool parameters, tool results, tool performance, and FinOps token usage.
Enable Secure Use of Vercel’s AI Gateway
Vercel’s built-in AI Gateway is fantastic for managing token counts, tracking latency, and optimizing model routing. However, model access and threat security are not the same thing.
Vercel’s gateway lacks native, deep-packet inspection for LLM threats, prompt injections, data exfiltration, and/or malicious tool manipulation. It also doesn’t currently allow your team the ability to natively deploy third-party security guardrails directly into the agent runtime.
Without a dedicated AI security solution, a compromised agent or an untrusted user prompt can weaponize a Vercel Managed Connector to exfiltrate database records from Snowflake or drop malicious commits into GitHub. Vulnerability types the Noma Labs team has already detected and reported in the wild.
Enter Noma: Native AIDR for the Vercel AI SDK
Noma bridges this gap by seamlessly overlaying enterprise AI Detection and Response (AI-DR) onto Vercel Services.
Because the Vercel AI SDK is modular and can be configured with any AI Gateway, deploying Noma is incredibly straightforward. By routing your Vercel AI SDK traffic through any Noma-supported gateway architecture (such as Kong or LiteLLM), you instantly unlock full Noma AIDR protection.
How Noma Enhances Vercel Agent Security:
- True Third-Party Guardrails: While Vercel tracks what the agent is doing via OpenTelemetry, Noma evaluates the intent and safety of the prompt, the model’s reasoning, and the tool outputs in real time, blocking threats before they hit your execution sandboxes.
- Context-Aware Tool Protection: EVE allows human-in-the-loop approvals for destructive tools. Noma takes this further by autonomously analyzing the parameters passed to Vercel Connectors (like Salesforce or GitHub). If an agent attempts an anomalous or unauthorized tool call, Noma flags or blocks it instantly.
- Advanced Prompt Injection & Exfiltration Defense: Noma ensures that data passing through Vercel Managed Connectors isn’t leaked via clever indirect prompt injections hidden within third-party data sources.
Secure Your Agent Infrastructure Today
Vercel has built the ultimate launchpad for the agent era, eliminating the friction of full-stack deployment. But to confidently run autonomous agents in production, enterprise teams need a security architecture that can not only keep up, but stay ahead of what’s coming.
By pairing the developer agility of Vercel Services and the EVE framework with Noma’s AI Detection and Response solution, you get the best of both worlds: unprecedented agentic power and uncompromised enterprise AI security.
Ready to secure your Vercel AI SDK deployments? Reach out to Noma today to see our AIDR solution in action and begin protecting your Vercel AI Stack.
