Vercel ships Connect, eve, and Passport in a three-launch push past hosting

Vercel shipped three launches on 2026-06-17 that read as a single thesis. Vercel Connect entered public beta as a runtime credential broker, eve opened on GitHub as an open-source agent framework, and Vercel Passport reached public beta as a deployment-level access gate. The throughline is Vercel staking the agent-runtime, secret-broker, and identity layers as platform surfaces in their own right rather than dependencies the user is expected to assemble.

What shipped

Vercel Connect replaces the long-lived secret in an environment variable with a runtime token exchange. An application proves its identity via OIDC and receives a short-lived, task-scoped token from the configured provider (Slack, GitHub, Linear, and a growing list at launch). Hobby includes 5,000 token requests per month at no cost; Pro and Enterprise meter at $3 per 10,000 requests. The audit story is what the pricing buys: every token mint is traceable to the calling function, which is the change that makes agent runs auditable in a way that “we set OPENAI_API_KEY in the env var” never was.

Eve is the framework most directly aimed at the LangChain, Mastra, and OpenAI Agents SDK slots. Built-in durable execution, sandboxing, and observability are the headline differentiators, and the deploy target at launch is Vercel Functions, with the project noting that “support for other platforms” is on the way. Local development runs against Docker, microsandbox, or just-bash through an adapter pattern, and the model layer stays open via OpenAI-compatible APIs. The launch blog positions eve next to Next.js in scope, which is the deliberate framing of “we did this for the web, now we’re doing it for agents.” Open-sourcing at github.com/vercel/eve from day one is the part that matters operationally; the framework’s success depends on Vercel not being the only deployment target a year out.

Vercel Passport reaches public beta as a $100-per-project monthly deployment gate that fronts apps with Okta, Auth0, or any OIDC-compliant identity provider. The use case is the deployment that needs SSO before any rendering happens, which up to now meant either a custom Edge Middleware shim or a third-party tunnel. Passport collapses that into a project setting.

Where this lands in the market

The three launches map cleanly onto categories Vercel was previously content to integrate around. Credential brokering was Auth0’s Token Vault story and Cloudflare’s MCP gateway story. Agent frameworks were a LangChain, Mastra, and OpenAI Agents SDK conversation. Deployment-level access was the Cloudflare Access lane. The combined message is that Vercel is no longer comfortable letting those layers live outside its surface area, because each one is a place where customers building agentic workloads on Vercel Functions might otherwise route through another vendor’s plane.

Connect is the most defensible of the three. The token-exchange architecture matches the direction Anthropic, Cloudflare, and the broader OIDC-native ecosystem are pushing, and Vercel’s first-party visibility into the calling function is the integration the others cannot match without identical execution context. Eve is the riskier bet because it walks into a category that already has product-market fit holders, and the open-source distribution model means Vercel needs to win on developer ergonomics rather than billing leverage. Passport is the most narrowly scoped: $100 per project per month is a deliberate price for the small-to-medium team that wants SSO without a Cloudflare contract.

The week’s earlier launches (Vercel Drop on 2026-06-12, the Grok Build plugin on 2026-06-11, and the GLM 5.2 AI Gateway addition on 2026-06-16) now read as the supporting context for the platform thesis. Drop intercepted the AI-builder export. Grok Build extended the gateway story. The Wednesday cluster names the three layers Vercel intends to own across the agent lifecycle.

What’s worth watching

1. Eve adoption outside Vercel. The open-source distribution model means eve is judged on third-party deploy targets, not Vercel Functions installs. If serverless adapters for Cloudflare Workers, AWS Lambda, or self-hosted Node land within two quarters, eve is a credible cross-platform framework. If they don’t, eve is a Vercel-specific agent runtime with an open-source license.
2. Connect token economics. Token-exchange pricing at $3 per 10,000 requests reads cheap for credential brokering and expensive for high-frequency agent loops. The first customers hitting Pro-tier ceilings will tell whether the meter is the binding cost in agent runs or whether the underlying inference still dominates.
3. Passport pricing collision with Cloudflare Access. $100 per project per month against Cloudflare Access’s per-user pricing puts the two on different cost curves at different team sizes. Watch for whether Vercel ships volume-based tiers as Passport leaves beta or holds the flat-rate framing as an explicit positioning move.

The plain reading is that Vercel is not adding products, it is consolidating layers. Connect, eve, and Passport are each in categories where Vercel was previously a willing integrator; the launch day signals the company has decided those categories are platform surface, not partner surface. Whether the platform survives the increased fight that posture invites depends on how quickly Connect’s calling-context advantage, eve’s ergonomics, and Passport’s price point hold up against the incumbents now contesting them.

Sources cited

1. Vercel: Introducing Vercel Connect vercel.com
2. Vercel: Introducing eve vercel.com
3. Vercel: Vercel Passport is now in Public Beta vercel.com
4. GitHub: vercel/eve github.com