Vercel’s
next chapter.

Your Python backend can share the same home as your frontend.

Python on Vercel is now generally available. FastAPI, Flask, and Django get automatic framework detection, with standard CPython, PyPI packages, and support for uv and pyproject.toml.

Vercel Services brings the frontend and backend into one project with shared previews, deployments, and rollbacks. Native SDKs cover models, storage, queues, and more; the AI SDK and Workflow SDK for Python remain in beta.

Meet Daemons: background workers that stay running between jobs.

Now in public beta, Daemons runs containerized processes that pull work from queues, databases, or other systems. You choose the instance count in each region; Vercel maintains it and restarts workers that exit or crash.

That fits Celery workers, Kafka consumers, Temporal workers, and CI runners. Launch support focuses on pull-based work, so these workers don’t need an inbound HTTP server.

Connect is the bridge to the tools and data an agent needs.

Vercel Connect handles provider credentials and OAuth flows so your app can request scoped access at runtime. It connects apps and agents to services such as Slack, GitHub, Linear, and Snowflake.

New today: Secure Compute routing lets supported connectors obtain and refresh tokens through a selected network’s dedicated IPs. That helps with providers that require an IP allowlist. Your app’s direct API traffic is unchanged by this setting.

Token Pass bundles credit for a selection of models through AI Gateway.

At launch, Pro teams can join early access for $30 per month, with $80 in monthly credit for GLM 5.3 and Kimi K3. The team shares that credit across compatible coding tools or a personal agent built with the AI SDK.

Unused credit doesn’t roll over. Requests stop when it runs out, with no automatic overages. The subscription renews monthly unless canceled, and Zero Data Retention is enabled by default.

Launch terms checked October 15. This covers the included models, not every model on AI Gateway.

Drives gives an agent a workspace that lasts beyond one run.

Vercel Drives is now generally available for Sandbox on all plans. Create a drive once, then mount it into a new sandbox when the next task needs a different environment.

Agents can reuse prepared data, mount read-only copies, or fork a drive into separate writable workspaces. Each fork can explore its own approach; forks don’t merge back into the original.

An open-source framework that gives the agent a structure.

Introduced June 17 and in public beta, eve brings the Agent Stack together. Instructions live in Markdown, tools in TypeScript, and the agent’s skills, channels, and schedules live alongside them in a directory.

Durable sessions, sandboxed compute, human approvals, subagents, and evals are built in. The appeal is spending more time defining the work the agent should do.