Overalls Launches Self-Hosted AI Developer Platform

Overview

Overalls, a new entrant in the developer tooling space, has officially launched its self-hosted AI developer platform. Designed for organizations that require strict data sovereignty and infrastructure control, the platform allows teams to deploy persistent, AI-ready development environments on their own hardware.

The service positions itself as an alternative to managed cloud IDEs like Gitpod or GitHub Codespaces, but with a critical distinction: everything runs on infrastructure owned and operated by the user. By combining persistent virtual machines with built-in AI agent capabilities, Overalls aims to streamline the software development lifecycle while keeping sensitive code and API keys off third-party servers.

The platform is open source under the MIT license, meaning there is no hosted version to sign up for. Instead, organizations install and run the software themselves, ensuring that no customer material leaves the network and that all development activity remains within their controlled perimeter.

Core Architecture and Machine Management

At the heart of the Overalls platform is a robust machine management system built on Firecracker microVMs. Each virtual machine operates with its own isolated kernel and a ZFS disk, providing a secure and efficient environment for development tasks. A key feature of this architecture is persistence; when a machine is stopped, its disk, hostname, and network address are preserved, allowing developers to resume work exactly where they left off.

Users can perform a variety of lifecycle operations on these machines, including cloning, resizing, and migration. It is possible to resize CPU and memory allocations or grow the disk up to 500 GiB. Stopped machines can also be moved between physical hosts without losing their identity, offering significant flexibility in resource management.

Operating System Support

The platform supports multiple guest operating systems, catering to diverse development needs:

NixOS: The default guest environment, chosen for its declarative and reproducible nature. It comes pre-configured with Nix and devenv, allowing for per-project environment isolation. Images are built from pinned package snapshots, ensuring that every build is identical.
Ubuntu: A widely used Linux distribution supported for general development workflows.
Windows: Added support for teams requiring Windows-based toolchains and applications.

Additionally, the company is actively developing support for RHEL-compatible guests, FreeBSD, and illumos. These additions address specific niches, such as testing BSD software on a real kernel rather than through emulation, or working on Solaris-heritage systems.

Pre-installed Tooling and Network Security

Linux guests come equipped with a comprehensive suite of developer tools out of the box. This includes Podman (compatible with Docker), git, the GitHub CLI, tmux, Neovim, ripgrep, jq, and direnv. For package management, the platform utilizes a tool called Brady. Brady acts as a secure proxy for downloads from sources like apt, Nix, pip, and npm, fetching only from verified upstreams and caching responses. This ensures that subsequent installations are fast and that no unverified software enters the environment.

Security is paramount in the design. The platform explicitly supports configurations with no outbound network access, no external sharing, and no web exposure. SSH access is restricted to the owner, ensuring that sensitive projects remain isolated from external threats.

AI Integration and Agent Workflows

One of the most significant aspects of Overalls is its integrated approach to AI-assisted development. The platform comes with popular AI coding tools such as Claude Code and Codex pre-installed on compatible guest images. However, unlike traditional setups, no API keys or credentials are stored on the machines themselves.

The Ziggy Proxy

To manage AI inference securely, Overalls uses a component called Ziggy. All requests from guests to OpenAI or Anthropic models are routed through an internal endpoint at https://llm.internal. Ziggy handles the authentication with external providers, meaning that sensitive API keys never touch the virtual machine.

This architecture allows for granular control over usage. Every token consumed by an AI agent is accounted for and recorded. Users can monitor their own consumption down to the specific machine and agent thread, while administrators have access to fleet-wide reports. These reports detail AI adoption, utilization rates, and estimated costs, which can be exported to reconcile against bills from inference providers.

Persistent AI Agents

The platform features a built-in agent workflow that mirrors the functionality of Amp orbs but operates on hardware the user owns. Developers can initiate a thread on any machine directly from the browser or via the CLI. If the target machine is powered down, the agent will automatically wake it to continue its work. This ensures that long-running development tasks, such as complex refactors or extensive test suites, can proceed autonomously without requiring a developer to keep a laptop open.

Collaboration and Developer Experience

Overalls is designed to facilitate seamless collaboration among distributed teams. Access control is managed through roles defined in Cedar, a policy language that checks every action in real-time. The available roles include:

Viewer: Can watch the terminal session but cannot interact.
Operator: Can take control of the terminal and hand it back.
Manager: Can start, stop, delete, and re-share the machine.

The owner of a machine cannot be removed, ensuring that final authority remains with the creator. When access is revoked, Cedar ensures that the change takes effect immediately, even if pages are already open in browsers.

Cross-Platform Client and Browser Integration

The overalls client is available for macOS, Linux (x86_64 and arm64), and Windows. Installation is streamlined with a one-line setup that configures SSH, allowing users to connect to any machine regardless of its operating system. The CLI supports standard commands such as ssh, ls, start, stop, and rm, with JSON output and retry-safe request IDs for scripting purposes.

For browser-based work, each machine can serve HTTPS on a subdomain under the user's domain. Additionally, Overalls offers a personal code editor built from Microsoft's official Code OSS source. This editor runs as the user inside the guest and supports Open VSX extensions, with verification currently focused on NixOS.

Git Integration and Workflow

The platform supports a workflow similar to Gitpod, allowing developers to spin up environments directly from repositories. By appending a GitHub URL to the Overalls domain, users can create a new machine with the repository cloned into /workspace/. This supports specific branches, commits, or pull requests.

A browser extension enables one-click initialization from GitHub, GitLab, Codeberg, or Bitbucket Cloud. The extension sends only the page's origin and path, preserving privacy while providing immediate access to the relevant codebase. This encourages a "one machine per piece of work" mentality, where environments are deleted after the task is complete, reducing technical debt and environment drift.

Strategic Implications

The launch of Overalls reflects a growing trend among enterprises to reclaim control over their development infrastructure. As AI tools become integral to the coding process, the risk of exposing proprietary code to third-party LLM providers has heightened concerns about data leakage and compliance.

By offering a self-hosted solution that pre-installs major AI agents but isolates API credentials, Overalls provides a middle ground. It allows teams to leverage the productivity gains of AI coding assistants while maintaining the security guarantees required by regulated industries. The open-source nature of the project also invites community contributions and customization, potentially accelerating its adoption in environments with unique technical requirements.