Vercel used its Ship 2026 conference to consolidate a year of AI infrastructure releases into a single platform called the Agent Stack, combining model routing, durable execution, sandboxed compute, and a new credential management layer. Alongside it, the company open-sourced eve, an agent framework built on the stack, and launched Vercel Connect in public beta.

The announcements position Vercel as a full-stack agent deployment platform, not just a frontend hosting company.

What the Agent Stack Includes

The Agent Stack bundles five existing Vercel products with two new ones. AI SDK provides a single interface for calling any model. AI Gateway routes calls across hundreds of models from one endpoint with automatic failover. Workflow SDK checkpoints every step of every agent job so runs resume from the last good step instead of restarting from zero. Vercel Sandbox gives each agent its own isolated microVM with a filesystem, Docker support, and its own kernel.

The new additions are Vercel Connect and Chat SDK, which delivers agents across Slack, GitHub, Linear, WhatsApp, Discord, and other platforms from a single codebase.

“The last thing we want is to rebuild our infrastructure every time a new model drops,” Greg Chan, CTO of real estate firm SERHANT., told Vercel. SERHANT. runs three models from a single key through AI Gateway, routing market analysis to Claude, marketing copy to GPT, and image generation to Gemini.

Eve: An Opinionated Agent Framework

Eve structures agents as directories. An agent.ts file defines the model. An instructions.md file serves as the system prompt. Tools, skills, subagents, channels, and schedules each get their own files. Vercel describes it as “Next.js for agents,” according to its announcement.

The framework ships with durable execution, sandboxed compute, human-in-the-loop approvals, subagent delegation, tracing, and evals built in. Agents can connect to external services through MCP servers or any API with an OpenAPI document, with credentials brokered through Vercel Connect so the model never sees raw tokens.

Eve supports deployment across multiple channels through small adapter files: HTTP, Slack, Discord, Teams, Telegram, Twilio, GitHub, and Linear are included at launch.

Connect Replaces Long-Lived Tokens

Vercel Connect replaces stored API tokens with runtime credential exchange. Developers register a connector once, then the agent mints a short-lived token scoped to each specific task at runtime. Every action traces from user to agent to service in an audit log.

At public beta launch, Connect supports Slack, GitHub, Snowflake, Salesforce, Notion, and Linear, plus any service reachable over OAuth or API key. Tokens can be scoped per-user, so an agent acting on behalf of different users gets credentials tied to each individual’s authorization rather than a shared bot token.

Revocation works at the connector level. Developers can revoke their own tokens or all tokens across every user and installation with a single CLI command, according to the Connect documentation.

Token Volume Context

Vercel’s AI Gateway production index for June 2026 reported that the gateway grew from approximately 2 trillion to 20 trillion tokens per month year-over-year, a 10x increase. Anthropic held 65% of all spend in May 2026, while DeepSeek V4 captured 17% of token volume in its first month of availability, at roughly 1% of spend. The index covers hundreds of models across production applications.

Consolidation in a Fragmented Market

The Ship 2026 announcements arrive in a week where Databricks released Omnigent for multi-agent coordination, TrueFoundry launched Agent Gateway, and Akamai debuted its Know Your Agent protocol. Vercel’s bet is that unifying model routing, execution, credential management, and deployment into a single platform reduces the integration burden that has defined agent infrastructure so far.

Eve is open-source and available at eve.dev. Vercel Connect is in public beta.