---
title: "Introducing Run SDK: secure eval for your agents"
url: https://daily.dev/posts/introducing-run-sdk-secure-eval-for-your-agents-iifyzzljd
source_url: https://vercel.com/blog/introducing-run
type: article
source: "Vercel"
published: 2026-08-25T19:21:24.011Z
updated: 2026-08-25T19:42:49.296Z
tags: ["ai-agents", "typescript", "vercel"]
reading_time: 5
upvotes: 1
comments: 0
language: en
---

> ## Documentation Index
> Fetch the complete documentation index at: https://daily.dev/llms.txt
> Use this file to discover all available pages before exploring further.

# Introducing Run SDK: secure eval for your agents

**[Vercel](https://daily.dev/sources/vercel)** · 5 min read · 1 upvotes · 0 comments

## Summary

Vercel released the Run SDK, a package for executing untrusted JavaScript and TypeScript without giving the code direct access to the host application or system. It runs code in a fresh QuickJS context inside a worker thread, exposing only explicitly defined host functions, and can interrupt execution mid-run for human approval or authentication before resuming without redoing completed work. The SDK is the internal module powering code mode tool execution in the AI SDK, supports Node.js 22.13+ and Bun, and is intended for JavaScript computation rather than OS-level or process isolation workloads (for which Vercel Sandbox is recommended).

## Full article

daily.dev links to this article rather than hosting it. Read it at the original source: <https://vercel.com/blog/introducing-run>

## Questions this post answers

### What is the Run SDK from Vercel and what does it do?

The Run SDK is a package for executing untrusted JavaScript or TypeScript without giving it direct access to an application or system. It runs code in a fresh QuickJS context inside a worker thread with no direct route to Node.js or the network, exposing only explicit host functions defined by the application, installable via pnpm add run.

_Teams sandboxing agent-generated code can track new SDKs like this one on daily.dev._

### How does the Run SDK support human-in-the-loop approval for sensitive operations?

A host function can call context.interrupt() to pause execution mid-run, returning a signed token the application saves with an approval request; when a decision arrives, the run resumes from that point. Settled host function calls use their recorded results, so completed work does not repeat, and the interrupted function receives the approval to continue.

_Developers designing approval gates for agent workflows can follow patterns like this via daily.dev._

### What is code mode in the AI SDK and how does it relate to the Run SDK?

Code mode in the AI SDK lets a model generate a single TypeScript program to coordinate multiple tool calls and logic instead of issuing them one at a time, and it is powered internally by the Run SDK. Existing AI SDK tools are mapped to host functions, while the Run SDK handles their sandboxed execution inside a QuickJS worker thread.

_Anyone evaluating agent tool-calling approaches can compare code mode setups on daily.dev._

## Similar posts on daily.dev

- [Vercel Sandbox tutorial: run untrusted code safely](https://daily.dev/posts/vercel-sandbox-tutorial-run-untrusted-code-safely-ycdiv657b) · Flavio Copes · 1 upvotes · 1 comments

---

Tags: [#ai-agents](https://daily.dev/tags/ai-agents), [#typescript](https://daily.dev/tags/typescript), [#vercel](https://daily.dev/tags/vercel)

[View this post on daily.dev](https://daily.dev/posts/introducing-run-sdk-secure-eval-for-your-agents-iifyzzljd)