EveryDev.ai
Subscribe
Main Menu
  • Tools
  • Developers
  • Topics
  • Discussions
  • Communities
  • Users
  • News
  • Podcasts
  • Blogs
  • Builds
  • Contests
  • Compare
  • Arena
  • Polls
Rate toolsCreate
AI Tools by Topic
  • AI Coding Assistants
  • Agent Frameworks
  • MCP Servers
  • AI Prompt Tools
  • Vibe Coding Tools
  • AI Design Tools
  • AI Database Tools
  • AI Website Builders
  • AI Testing Tools
  • LLM Evaluations
Follow Us
  • X / Twitter
  • LinkedIn
  • Reddit
  • Discord
  • Threads
  • Bluesky
  • Mastodon
  • YouTube
  • GitHub
  • Instagram
Get Started
  • Users
  • Rate Tools
  • About
  • Editorial Standards
  • Corrections & Disclosures
  • Community Guidelines
  • Advertise
  • Contact Us
  • Newsletter
  • Submit a Tool
  • Start a Discussion
  • Write A Blog
  • Share A Build
  • Terms of Service
  • Privacy Policy
Explore with AI
  • ChatGPT
  • Gemini
  • Claude
  • Grok
  • Perplexity
Agent Experience
  • llms.txt
Theme
With AI, Everyone is a Dev. EveryDev.ai ©
    1. Home
    2. News
    3. Vercel Ship 26: Agents Get a Login, Event Triggers, and a Budget

    Vercel Ship 26: Agents Get a Login, Event Triggers, and a Budget

    Mateo Rivera's avatarMR
    Mateo Rivera
    November 18, 2026·Open-source tinkerer exploring local…
    Discuss (0)
    Vercel Ship 26: Agents Get a Login, Event Triggers, and a Budget

    Deployments triggered by agents on Vercel grew 28-fold over the past 12 months, CEO Guillermo Rauch said at Vercel Ship 26 in San Francisco on October 16.

    Most of the announcements addressed what happens after an agent starts running. It needs permission to act on someone’s behalf, a way to detect changes, persistent storage for its work, and limits on what it can spend.

    Here’s what Vercel shipped, who can use it, what it costs where pricing is available, and what I’d try first.

    What shipped and who can use it

    FeatureStatusCost
    Vercel ConnectGenerally available, with 1,000+ connectorsHobby includes 500 token requests and 1,000 trigger forwards; Pro costs $3 per 1,000 token requests and $0.95 per 1,000 forwards
    Enterprise-Managed AuthorizationPublic betaRequires a compatible identity provider, such as Okta
    MCP Events in ConnectPublic betaNot stated separately
    Drives for SandboxGenerally available on Hobby, Pro, and EnterpriseNot listed in the announcement
    eve persistent memoryAvailable in the open-source frameworkDepends on the memory provider
    Vercel DaemonsPublic betaNo pricing published
    AI Gateway budgets and US-only inferenceAvailableBudgets are soft caps
    Token PassEarly access for Pro teams$30 per month for $80 of GLM 5.3 and Kimi K3 credit
    PythonGenerally availableStandard Vercel compute pricing
    Bring Your Own Cloud on AWSGenerally availableEnterprise plans only

    Vercel Connect gives agents a login

    Suppose an agent needs to post in Slack and read a Linear issue on behalf of a user. Someone has to obtain an OAuth token for each service, store it securely, refresh it before it expires, and ensure the agent receives only the scopes it needs.

    The usual shortcut is a long-lived token stored in an environment variable. That is exactly the kind of secret you do not want an agent to be able to read.

    Vercel Connect handles this work. An application asks Connect for a token with specific scopes, and Connect manages the exchange, caching, and renewal. The service launched in public preview at Ship London and is now generally available, with more than 1,000 connectors spanning OAuth, MCP, and API-key services. Experimental connectors are marked in the catalog.

    Enterprise-Managed Authorization removes consent prompts

    The token problem has another layer inside a company. If an internal agent works across Slack, GitHub, and Notion, every employee would normally have to approve a separate connection for each service. IT would also lack a central place to revoke that access.

    Enterprise-Managed Authorization, now in public beta, moves the decision to the company’s identity provider. An employee signs in with a corporate account through a compatible provider such as Okta. An administrator can then grant or revoke access to supported services centrally.

    The application still requests specific scopes. The resulting token is limited both to those scopes and to what the employee could already do in the connected service.

    The underlying standard is Cross-App Access (XAA), an IETF draft in which an identity provider vouches for the user to another service. This removes the need for the user to approve every connection individually.

    Vercel’s example requests a token for Slack’s MCP server using the signed-in user’s identity:

    import 'server-only';
    
    import { getToken } from '@vercel/connect';
    import { requireIdentity } from '@vercel/passport';
    
    export async function getResourceToken(scopes: string[]) {
      const identity = await requireIdentity();
      return getToken('oauth/resource', { subject: identity, scopes });
    }
    

    The example uses Vercel Passport for identity, which makes this feature most relevant to internal enterprise applications rather than weekend projects.

    Connect also added Secure Compute routing. Services such as Salesforce and Snowflake often accept traffic only from allowlisted IP addresses. Connect can now route token requests through the static IP addresses of a Secure Compute network.

    If that network is unavailable, the requests fail instead of falling back to Connect’s shared IP addresses. That is the safer behavior for an IP allowlist.

    MCP Events let agents react instead of polling

    Most agents discover changes by repeatedly asking for them: Is there a new issue? A new message? Has anything changed since the last check?

    Polling consumes requests when nothing has happened and introduces delays when something has.

    MCP Events, now in public beta through Connect, reverses that model. An agent subscribes to an event on a compatible MCP server, such as the creation of an issue, and Connect delivers the event to the application as a webhook.

    Connect creates the subscription on the MCP server, renews it before it expires, verifies the server’s webhooks, and forwards each event through an authenticated delivery.

    import { experimental_subscribe } from '@vercel/connect/mcp';
    
    export async function subscribeToIssues(idempotencyKey: string) {
      return experimental_subscribe('oauth/issues', {
        tokenParams: { subject: { type: 'app' } },
        name: 'issue.created',
        arguments: { project_id: 'ABC' },
        destination: { path: '/hooks' },
        ttlMs: 24 * 60 * 60 * 1000,
        idempotencyKey,
      });
    }
    

    The receiving endpoint calls experimental_verifyMcpEvent to confirm that a delivery came from Connect before acting on it. The experimental_ prefix signals that the API may still change.

    These subscriptions are also available as agent tools in the AI SDK and eve, Vercel’s open-source agent framework. An agent can therefore discover the events exposed by a server and subscribe without a developer defining every subscription in advance.

    Support is also beginning to extend beyond Vercel. Cloudflare’s MCP server started sending Workers errors to subscribed agents through MCP Events this month.

    Drives and eve memory store different kinds of state

    Two announcements addressed persistent state, but they solve different problems.

    Drives are for files. A Drive is a directory mounted inside a Vercel Sandbox. Its contents remain available after the sandbox stops.

    A later run, or a different sandbox, can mount the same Drive and continue where the previous one stopped. An agent no longer needs to clone the same repository and reinstall the same dependencies every time it starts.

    eve memory is for information an agent learns. Memories live in named slots defined by files under agent/memory/. Each slot specifies a provider, which stores and retrieves the memory, and a scope, which determines who shares it. A slot might, for example, contain separate memories for each signed-in user.

    Before each turn, eve retrieves the relevant memories and adds them to the model’s context. Depending on the provider, memories can be updated automatically after a turn, explicitly through tools called by the agent, or both.

    When deployed, the built-in file provider uses a private Vercel Blob store, allowing memories to survive restarts and deployments. Supermemory, Upstash AgentKit, and Kybernesis Arcana are also supported, and developers can write their own providers.

    eve also gained extensions for installing tools and skills, along with a way to create and deploy agents from the Vercel dashboard.

    If the agent needs to…Use
    Preserve a checkout, build cache, or generated files between runsDrives
    Remember a user’s preference next weekeve memory

    Daemons and Python run the work behind an agent

    Not all agent work begins with an HTTP request. A worker might pull a job from a queue, process it, and immediately ask for another one. Until now, that kind of persistent process usually had to run somewhere other than Vercel.

    Vercel Daemons, now in public beta, are designed for these workloads. Developers can deploy code or a container image, choose how many instances should run in each region, and let Vercel restart instances that exit or crash.

    The initial release supports pull-based workloads such as queue consumers, database job pollers, Celery and Dramatiq workers, and Temporal workers. CPU and memory metrics require Observability Plus.

    I could not find published Daemons pricing on either the changelog or the product page. That is the number I would wait for before moving an existing worker fleet.

    Python also became generally available. FastAPI, Flask, and Django now run on standard CPython with PyPI dependencies, uv, and pyproject.toml. Vercel says 150,000 teams already run Python backends on its platform.

    For teams that write agent tools or MCP servers in Python, those services can now live beside the frontend in the same Vercel project.

    AI Gateway budgets are soft caps

    AI Gateway now supports spending budgets for a team, project, API key, or individual team member. After a budget is exceeded, the gateway rejects new requests that use Vercel’s system credentials.

    Two limitations matter if you plan to use these budgets to contain a runaway agent.

    First, Vercel describes them as soft caps. Requests already in flight can push spending beyond the configured limit.

    Second, usage through bring-your-own provider keys is metered separately and does not count toward these budgets. If an agent calls Anthropic or OpenAI through AI Gateway using your own provider key, the budget will not stop it.

    AI Gateway also added US-only inference, which keeps model processing within the United States, alongside its existing Zero Data Retention option.

    Token Pass is a new prepaid plan in early access. Pro teams pay $30 per month for $80 in credits for GLM 5.3 and Kimi K3. Unlike the standard budgets, Token Pass applies a hard stop once the credit runs out.

    Shopify’s engineers call themselves verification engineers

    Shopify CEO Tobi Lütke described how the company’s engineers work during the closing conversation with Rauch.

    “We now think of ourselves as verification engineers,” he said, “building harnesses and acceptance tests that make changes verifiable.”

    Shopify is rewriting its storefront renderer from Ruby to Rust and comparing the output byte for byte before switching over. At a company hackathon, its internal Slack agent, River, topped the code-deletion leaderboard.

    That idea connects the rest of the announcements. Connect, Events, Drives, and budgets all let agents act with less human supervision. They do not determine whether an agent’s work is correct.

    That verification still has to come from somewhere. Shopify’s answer is to build the tests first.

    What I’d try first

    1. Replace one polling loop. If an agent checks GitHub, Linear, or Slack on a timer, subscribe to the equivalent MCP event with a 24-hour TTL. Compare its reaction time and request volume with the polling version.

    2. Mount a Drive for one sandboxed agent. Choose an agent that clones a repository and installs dependencies on every run. Measure its startup time before and after adding the Drive.

    3. Put a budget on every API key assigned to an agent. Then audit whether any calls use your own provider keys, because that spending will bypass the budget.

    Sources

    • Vercel Ship 2026 recap, Vercel
    • Vercel Connect, Vercel
    • Vercel Connect adds Enterprise-Managed Authorization, Vercel
    • Vercel Connect now supports Secure Compute, Vercel
    • MCP Events support in Vercel Connect is now in public beta, Vercel
    • Vercel Drives is now generally available, Vercel
    • Persistent memory for eve agents, Vercel
    • Backends on Vercel: Python, containers, and Daemons, Vercel
    • Vercel Daemons is now in public beta, Vercel
    • Vercel AI Gateway documentation, Vercel
    • Introducing Token Pass, Vercel
    • Bring Your Own Cloud on AWS is now generally available, Vercel
    • Identity Assertion Authorization Grant (Cross-App Access), IETF draft
    • Cloudflare AI changelog, Cloudflare
    View tool: ChatGPT Codex
    Promoted

    Sponsored

    ChatGPT Codex

    ChatGPT Codex

    OpenAI's agentic coding tool that runs in ChatGPT, your IDE, and the terminal to handle complex engineering tasks like bug fixes, refactors, and PR reviews.

    View tool

    Comments

    Checking profile…