discobox-ai
Discobox is an open-source harness runtime that runs coding agents in isolated sandboxes with their own copy of a repository, desktop, browser, tools and services. It is designed to let developers run parallel agent sessions and build software factories while keeping credentials outside the sandbox and routing outbound traffic through an auditable policy proxy.
At a Glance
- Software developers and engineering teams
- Teams using AI coding agents such as Claude Code, Codex and OpenCode
- Organizations needing isolated, auditable agent execution and credential controls
- Engineering organizations automating issue triage, delivery, testing and other software-factory steps
- +1 more
AI Tools by discobox-ai
(1)Discobox
Sandbox Runtime for Coding Agents
Discussions
No discussions yet
Be the first to start a discussion about discobox-ai
Latest News
Discobox introduced publicly as an open-source isolated machine/sandbox for coding agents
Obot published “Meet Discobox: Give Your Coding Agent Its Own Machine,” detailing the product, security model and software-factory roadmap
Discobox v0.19.0 released
Discobox v0.19.1 released
Products & Services
Open-source client/server runtime for creating and managing isolated agent boxes. It clones a repository or copies a local directory into a sandbox, runs a selected coding harness, and returns work through Git commits, branches or pull requests. The project is distributed through Homebrew and install scripts for macOS, Linux and Windows/WSL; the source repository is licensed Apache-2.0.
Market Position
Discobox positions itself as an open-source, harness-agnostic runtime for giving coding agents a complete isolated machine rather than merely restricting processes or tool calls. Its differentiators are local and self-managed deployment, persistent and parallel boxes, Git-native work transfer, browser/desktop access, provider portability, and an external proxy/control plane that keeps credentials outside the sandbox. It competes for agent-sandbox and isolated coding-environment use cases with providers and platforms such as E2B, Daytona, Modal, Vercel/Cloudflare sandbox products, and other disposable development-environment tools; Discobox emphasizes the full engineering workflow and host-controlled security policy.
Leadership
Founders
Darren Shepherd
Obot AI Chief Architect and the builder/creator of Discobox. Obot describes its team as having prior open-source infrastructure experience at Cloud.com, acquired by Citrix, and Rancher Labs, acquired by SUSE.
Executive Team
Darren Shepherd
Creator of Discobox; Chief Architect, Obot AI
Obot's Chief Architect who built Discobox from the need to give coding agents isolated machines; he presented the project publicly in September 2026.
Founding Story
Discobox grew out of Darren Shepherd's frustration with giving coding agents access to his work laptop. He wanted agents to run applications, open browsers, test their work and install tools freely, while keeping the host machine and its credentials protected; the resulting vision was to give every coding agent its own isolated machine.
Business Model
Revenue Model
Discobox is presented as open-source software that users install and run on their own desktop, server or chosen infrastructure/provider. The public site and documentation do not state a paid subscription, hosted-service charge or other Discobox-specific monetization model.
Target Markets
- Software developers and engineering teams
- Teams using AI coding agents such as Claude Code, Codex and OpenCode
- Organizations needing isolated, auditable agent execution and credential controls
- Engineering organizations automating issue triage, delivery, testing and other software-factory steps
- Self-hosted and cloud infrastructure operators providing sandbox capacity
- Running coding agents safely with full machine access
- Parallel feature work, bug fixes and experiments against one repository
- Browser-based application testing and service execution by agents
- Persistent agents for triage, investigations and recurring engineering work
- Multi-agent software factories that use skills and a manager box to reconcile GitHub issues and pull requests
- Remote development and agent workloads on desktop, cloud or self-managed infrastructure