DEVOPSdigest

Incredibuild announced the launch of Islo, a sandbox for AI agents with granular security controls that allows engineering teams to safely run coding agents 24/7.

Islo provides a secure, isolated execution environment designed to manage how AI agents run, operate, and scale across the software development lifecycle.

Islo provides fully isolated execution environments with granular policy control, allowing organizations to run AI agents efficiently, confidently, and without disruption to existing development flows. With Islo, each AI coding agent gets its own dedicated, isolated environment independent of a developer's local machine, which can be operated according to enterprise policies and securely connect to critical systems such as GitHub and Jira, without exposing secrets or sensitive data. Developers can deploy agents into these isolated, cloud-based environments where every action is governed by explicit policies, controlling how agents access codebases, infrastructure, and external systems. With agents running in such isolated cloud environments, Islo decouples AI execution from local machines, enabling agents to operate 24/7, turning development workflows into continuous, autonomous systems with human checkpoints deployed as required.

This allows teams to move from ad hoc experimentation to controlled, production-grade AI workflows.

Key capabilities include:

- Isolated execution environments that prevent unintended access to sensitive systems

- Granular policy controls governing how agents interact with codebases, infrastructure, and external systems

- Reproducible environments that ensure consistent execution across runs

- Persistent execution contexts, allowing agents to operate continuously without repeated setup

- Integrated acceleration, leveraging Incredibuild's technology to speed up compute-heavy workflows

While Islo can be used independently, combining it with Incredibuild's acceleration capabilities can radically speed up AI-driven software development and reduce the time it takes to execute build, test, and compute-heavy CI/CD steps. This removes potential bottlenecks and allows teams to ship code at the velocity of AI.

"We've spent years helping teams ship quickly; Islo is making sure AI can ship safely," said Shimon Hason, CEO of Incredibuild. "We're introducing a missing layer in the stack: a super-powered sandbox that provides the infrastructure necessary for organizations to safely run AI agents as part of real production workflows."

"Coding agents are capable of doing real work now, but they all run on the developer's laptop. That means they die when the lid closes, and they have access to everything on the machine," said Adam Gold, Director of Product Engineering at Incredibuild. "We built Islo because we believe that every AI agent needs its own computer: Not an ephemeral container, but a long-running dev environment with its own running services, scoped credentials, and a lifecycle that doesn't depend on human supervision."

With Islo, Incredibuild continues to address the evolving needs of software development teams, offering a solution that supports AI integration while ensuring safety and performance in complex environments.

The same isolation and policy infrastructure that makes coding agents safe also addresses a fast-growing need in AI research: Running the large-scale evaluations, benchmarks, and reinforcement learning rollouts that modern agent development depends on. To serve this use case, Islo is partnering with the Harbor Framework community – an open-source infrastructure for authoring and executing agent benchmarks and evals. Through native Harbor integration, benchmark authors and RL engineers can run their tasks on Islo's cloud sandboxes with a single configuration change, gaining reproducible environments, reward-hacking defense, massively parallel execution, and deep observability in every rollout.