Marketing Team eve Template

The Marketing Team eve Template is an open-source project from Vercel Labs that lets you run a coordinated team of five AI marketing specialists directly from Slack. Built on the eve framework—Vercel's filesystem-first framework for durable backend agents—it routes work from a single lead agent to specialists that write long-form content into Notion, queue social posts in Typefully, build and send email campaigns in Resend, and audit pages for SEO. The template is licensed under MIT and published on GitHub with one-click deployment to Vercel.

What It Is

The template implements a multi-agent system structured as a lead plus five specialists. The lead agent never writes deliverables itself; it grounds each request in a shared brand context document and per-user preferences, then delegates to exactly one specialist with a complete brief. Each specialist runs in a fresh child session with no inherited conversation history, tools, or connections from the lead. The five specialists are:

• product-marketer — owns positioning, messaging, competitive alternatives, and the shared brand context document
• content-marketer — writes long-form blog posts, landing pages, case studies, and newsletters, delivering a Notion page link
• social-media-coordinator — drafts short-form posts for X, LinkedIn, Threads, Bluesky, and Mastodon via the Typefully queue
• seo — runs page and site audits, hierarchy and internal linking, JSON-LD schema, and templated page sets
• email — reworks copy for inbox fit, then builds and sends campaigns in Resend

How the Agent Architecture Works

The tree is deliberately one level deep. The lead delegates; specialists do not sub-delegate further. Each specialist loads skills on demand—per-platform style guides with banned-words lists, SEO audit checklists, email deliverability rules—using eve's load_skill mechanism. Skills add instructions to the turn but never add new actions the agent can take.

Specialists gather their own evidence using web_search and web_fetch against a source budget, run lint_against_style on each draft to catch banned words, and edit in deliberate passes against a shared writing-quality rubric before returning work. Long documents move between specialists by artifact ID rather than by pasting text through the lead's context window, keeping handoffs clean.

Twenty-three skills are distributed across the five specialists, and a writing-quality skill is shared via a defineSkill factory pattern rather than duplicated directories.

Human-in-the-Loop Approval Gates

Every irreversible action pauses for an approve-or-deny decision, rendered as buttons in Slack or as a prompt in the dev TUI. Sends in Resend are always gated—scheduled or not—because email cannot be recalled. Typefully deletes always pause, but creating or editing a draft only pauses when the call actually publishes or schedules it, keeping routine drafting friction-free. The Resend connection also narrows what the model can discover at all via a tools.allow list, reducing roughly 85 published Resend MCP tools to the 47 covering campaigns, lists, and diagnostics; account administration such as API key creation is never exposed.

Credential and Authentication Model

Authentication is handled at runtime with no shared integration tokens for most services. Notion, Resend, and Slack authenticate through Vercel Connect with per-user OAuth, so each action is attributed to the real signed-in person. Vercel Blob and the AI Gateway authenticate via the project's OIDC token. The only static credential in the project is TYPEFULLY_API_KEY, because Typefully's MCP server only accepts a static key.

User preferences are stored in Vercel Blob keyed by a SHA-256 hash of the framework-resolved principal ID, so a session can only read or write its own user's file and no raw identifier is stored in the path.

Setup and Deployment Path

The template deploys with a single Vercel button click that provisions Notion, Resend, and Slack connectors, a public Vercel Blob store, and prompts for the Typefully API key. One manual step remains after deploy: verifying a Resend sending domain in the Resend dashboard, which is skippable if email is not needed.

Local development requires Node.js 24 or newer, pnpm, and the Vercel CLI. After cloning the repository created by the one-click flow, vercel link and vercel env pull wire up environment variables including a short-lived OIDC token. The dev TUI (pnpm dev) provides a full local interface including approval gate prompts; the Slack surface cannot be tested locally because Slack forwards events through Vercel Connect to the deployed project.

Specialists can also be added as remote agents in separate deployments using eve's defineRemoteAgent and vercelOidc() for deployment-to-deployment auth with no shared secret.

Current Status

The repository was created on July 26, 2026 and last pushed on August 3, 2026, indicating active early development. The project had 305 stars and 12 forks as of early August 2026 per GitHub metadata. It is part of the Vercel Labs organization and sits alongside related templates including the eve Sanity Copilot and eve Content Agent templates.