AgentsThe story, in brief

Vercel Sandbox now supports Devin Outposts

Devin agents can now run in Vercel's infrastructure without keeping a local worker online—session isolation, durable workflows, and network policies included.

Illustration of independent geometric mechanisms passing paper tasks along branching amber tracks.
AI agents and the coordination of work.AI illustration by KeyNews
The KeyNews take

Why it matters

Cognition's Devin agent platform now integrates with Vercel Sandbox for distributed execution. This removes the operational burden of managing local Outpost workers while maintaining agent state and enforcing infrastructure policies—a significant step toward making agent deployment production-ready for teams.

The key facts

13 to know
  1. Devin sessions execute in isolated Sandbox microvMs managed by Vercel

  2. No local Outpost worker required to stay online

  3. Cognition control plane handles inference and planning; Vercel handles session orchestration and command execution

  4. Features: session snapshots for state preservation, Sandbox firewall rules for network policies

  5. Available to Pro and Enterprise teams with Devin Outposts enabled

  6. Encrypted credentials auto-provisioned; no manual API key management

  7. Integration announced August 4, 2026

  8. Devin Outposts now execute in Vercel Sandbox microVMs with session isolation

  9. Control plane stays with Cognition; orchestration and execution run in Vercel project

  10. Durable workflow manages session lifecycle

  11. Encrypted credentials managed automatically—no manual token/API key handling

  12. Available to Pro and Enterprise Devin Outposts users

  13. Features: Sandbox snapshots for session state, Sandbox firewall rules for network policies

Go to the source

Vercel Blogvercel.com

Publisher excerpt: now supports . Each Devin session executes in its own isolated Sandbox microVM, with no local Outpost worker to keep online.Vercel SandboxDevin Outposts The Devin control plane stays with Cognition, where the agent loop handles inference and planning. Session orchestration and command execution run…
Read original report
Back to today's editionMore agents news

The wider picture

View all
Illustration of independent geometric mechanisms passing paper tasks along branching amber tracks.
AI illustration by KeyNews
Agents01

Known Systems AI spins out of Identity Digital to make AI agents accountable to their owners

As autonomous agents move into production, their provenance and accountability become critical. Known Systems AI is building identity and governance infrastructure for agents — a layer between agent frameworks and deployment that enterprises will need to manage liability and trust.

SiliconAngle
Illustration of a transparent lens revealing connected networks across layers of paper.
AI illustration by KeyNews
Agents02

Cisco Talos finds malware that puts its next move to a four-model vote

AI-powered autonomous malware represents a novel attack surface: distributed decision-making agents that operate without traditional C2 infrastructure, forcing security teams to think about agent reliability and exploitation differently.

SiliconAngle
Illustration of independent geometric mechanisms passing paper tasks along branching amber tracks.
AI illustration by KeyNews
Agents03

Rabbit’s new AI agent doesn’t need an R1 to run

Rabbit pivots from failed hardware to cloud-native agent infrastructure. OS3 represents a shift toward device-agnostic autonomous systems that can reason over local files and apps—a meaningful deployment vector for autonomous work.

The Verge AI