Announcement of Vercel’s Eve Framework
This week at the Ship event in London, Vercel unveiled an open source agent framework called eve, alongside new innovations such as Passport, aimed at managing employee applications generated by AI under corporate oversight. The buzz around agents is palpable, especially custom varieties. Although numerous agent frameworks are already available to simplify programming, eve offers some distinct features.
The Mechanics of Eve
Using TypeScript and Markdown, an agent comprises a folder filled with files that outline instructions, capabilities, model providers, tools, authentication, channels, and scheduling. By default, agents are assigned their own sandboxed VMs. The framework also includes a straightforward testing tool that evaluates the agent’s performance and outcomes. The source code can be found on GitHub licensed under Apache 2.0.
Simplicity as a Selling Point
While a multitude of agent frameworks exists, Vercel’s CTO Malte Ubl emphasizes that eve’s advantage lies in its simplicity, comparing it to a “fill in the blanks” method. “The agent’s life cycle is fully managed by the framework, and as a developer or creator, you need to position elements correctly, but then everything operates seamlessly,” Ubl stated. He reassures that it’s a setup where you needn’t dive into the details of sandboxes or compacting context windows.
Launching with Eve
Agents developed with eve are deployed on Vercel by default, utilizing the same easy command for web applications: vercel deploy. Nonetheless, the company insists that it’s not confined to its own platform. “We are entirely dedicated to ensuring it operates everywhere,” Ubl noted, although an early user reported a bug regarding the requirement of a Vercel login when intending to use a different model provider. This issue may be resolved shortly as it is still in early stages.
LLM Flexibility
Regarding the LLM eve utilizes, Ubl states, “You can connect any model supported by the AI SDK, which includes all the models,” with the SDK being part of Vercel. They also provide Vercel’s AI Gateway, which features a unified endpoint for multiple model providers, enhancing reliability by switching to another model when one fails.
Enterprise Applications and Agents
The company also teasingly showcased Enterprise Apps and Agents, consisting of four elements. Vercel Connect replaces static secret credentials with temporary tokens using OAuth or an API. Vercel Passport organizes all team apps and AI agents behind an identity provider like Okta or Microsoft Entra. Enterprise Managed Users synchronize directories so that Vercel within a team is directed by the organization’s identity framework. Finally, Bring Your Own Cloud (BYOC) allows companies to utilize Vercel’s platform on customer-provided AWS infrastructure.
Advantages of Passport
According to Vercel, Passport is a highly desirable feature due to the number of employees generating applications hosted by Vercel, but outside the control of their organization. A typical scenario involves an employee creating an app with AI assistance, which defaults to the Next.js React framework and Vercel hosting. This demonstrates a type of shadow IT – or shadow AI – where employees create applications using company data while bypassing the organization’s IT regulations or controls.
Analyzing Vercel Platform’s Strengths and Limitations
Being an AWS customer, Vercel’s platform should work seamlessly with BYOC, although there are drawbacks. One drawback Ubl pointed out, “we don’t permit your compute to adopt AWS roles… If you’re deeply entrenched in AWS IM [Identity Management] security, Vercel’s not suitable for you,” Ubl explained. However, OIDC tokens are provided for every compute invocation to establish AWS policies.
Resource Efficiency vs Expense
All Vercel deployments are immutable, signifying that “every time you push to Git, a new infrastructure is created from scratch,” Ubl remarked, viewing this as ideal for AI agents. However, a common complaint is that Vercel operating on AWS incurs higher hosting costs than what is available directly. Ubl asserts that Vercel’s resource efficiency “particularly at low scale and in comparison to Lambda” alleviates this additional cost. Last year, Vercel reported reducing Lambda expenses by 95 percent by reusing idle instances.
Competition with Rivals
Cloudflare stands out as another competitor to Vercel. Unlike Vercel, Cloudflare maintains its own data centers with an efficient serverless platform using Workers, built on V8 isolates, which is a feature of the V8 JavaScript engine utilized by Google Chrome and the open-source Chromium project. Ubl acknowledged that while Cloudflare Workers are distinctive, Vercel stands as a more conventional platform, operating Node.js, Python, or PHP on a VM. Vercel offers standard PostgreSQL, VPC peering, AWS, S3; nothing custom.
Wrap-up
A technical debate is heating up here. Cloudflare’s engineering director Steve Faulkner, in February, referred to Next.js tooling, supported by Vercel, as “entirely bespoke.” Significant strides have been made since then, with an Adapter API stabilizing in Next.js 16.2. This means other providers won’t need to reverse-engineer build outputs, although AWS and Cloudflare adapters are still being developed, with completion expected by the end of 2026.
Quick Recap: Yay or Nay?
Vercel’s eve and Passport launch is certainly causing a stir in the AI realm. Whether they are as valuable as they seem is still up for discussion. Yet, a bit of technical discussion always helps get the creative juices flowing. For everything tech-related, stay tuned to gadgetlad.co.uk.
