Skip to main content
AI Development Environment is a self-hosted dashboard that coordinates AI-assisted development across every machine you enroll. It groups repositories into apps and keeps their Git worktrees, AI Plans and Sessions, builds, shell commands, and durable workflows in one control plane, so long-running work survives restarts and stays inspectable.

A pass through the Action Center, Worktrees, a worktree's detail page, and Sessions.

The Worktrees page in that tour is a good example of how the product works. From a single view you can create a worktree from a branch or a Jira ticket, open it in VS Code, sync, rebase, push, stash, or move it to another Mac, and turn on Auto Sync and Auto Merge so branches keep themselves current and merge once required checks pass. Every enrolled agent’s checkouts appear side by side.

​
What you can do

Manage each app in one place

Group one or more repositories into an app. Open its hub to work with only that app’s repositories, worktrees, Plans, Sessions, and builds.

Run AI Plans and Sessions

Survey a codebase read-only with a Plan, then start a Session from it against a worktree. Steer live transcripts, answer questions, and review the resulting changes and cost.

Manage worktrees and codebases

Track repositories and checkouts across agents, manage branches and stashes, and share settings between every checkout of a repository.

Review changes with coverage

Read any worktree’s diff in a full-window reviewer — staged, unstaged, branch, or per commit — with a build’s code coverage painted onto the lines.

Build, sign, and ship iOS apps

Configure and run iOS builds, inspect logs, artifacts, tests, and coverage, then export or install over the air. Manage provisioning profiles and enroll devices with Apple.

Automate with workflows and commands

Draw durable automations on a typed canvas, publish them, and watch runs live. Save reusable shell commands that run through an enrolled agent’s login shell.

Stay on top of GitHub and Jira

Review pull requests and review threads, monitor Actions runs and webhook deliveries, and browse Jira projects, boards, and saved JQL queries.

Debug what is running

Search console output from builds and devices, watch analytics events arrive, symbolicate crash reports with your dSYMs, compose and send Apple push notifications, and read everything on one live timeline.

​
Focus on an agent

On the web, use Active Agent, immediately to the left of Search, to focus the agent filters on Worktrees, Usage, Commands, Workflows, Plans, and Sessions. The selector starts at None. Search by agent name or hostname, then choose any enrolled agent, including an offline one. The iOS app supports agent focus on Worktrees and Usage.
Active Agent menu with searchable agents in light themeActive Agent menu with searchable agents in dark theme
While an agent is active, each affected page’s Agent filter shows that agent, is disabled, and says Controlled by Active Agent. Other filters still apply. Select None in the global selector to restore each page’s previous agent choice. Worktrees views inside individual apps keep separate saved choices. The browser remembers your active agent and previous page choices across navigation and reloads. Tabs signed into the same account share the browser’s active choice. The iOS app remembers its own selection for each server and account; selecting an agent on one device does not change another device’s choice. On mobile web, the selector collapses to an icon with an active indicator. In the iOS app, tap the dedicated Active Agent toolbar icon on the home destination list or a feature page. On iPad, it appears beside Search. The searchable sheet includes None to restore page filters. An agent with no matching worktrees, usage, or runs produces empty results. Offline agents remain selectable. A failed catalog refresh preserves your selection; a successful refresh that confirms the agent was removed clears it. Pages without an agent filter keep their usual behavior, and active focus does not change where commands, builds, or other operations run.

​
How it fits together

  • The control plane is a Next.js app with a GraphQL API and a SQLite database that you run on your own machine or server.
  • Each managed machine runs a control agent that makes outbound HTTP and GraphQL WebSocket connections to the control plane. Managed machines and agent containers never expose a listening port.
  • Agent identity and job history are durable, so jobs, builds, and Sessions survive restarts while subscriptions deliver live output.
  • Long-lived Jira, GitHub, signing, and APNs credentials and their connection settings go through a server-only credential service backed by the database, HashiCorp Vault, or the macOS Keychain.

Quickstart

Install the control plane, enroll your first agent, and open the dashboard.