No. The terminal is one capability inside a control plane for projects, logs, ports, Git, tunnels, databases, and machine operations.

No. Projects and commands run on your registered machine. Explicit Git publishing and AI actions send only the content needed to the provider or repository you select.

No. Register the folders and commands you already use. Dev Server Deck is designed to extend existing workflows rather than replace them.

Yes. The Agent supports native project paths and WSL-aware workflows, with runtime and task discovery across the supported ecosystems. In Studio, the coding agent can read, write, run commands, and rename, delete, or create files and folders on a WSL-hosted project the same as a native one - the only piece still missing is live file-watching for changes made outside Studio itself.

Yes. A one-line installer sets up a systemd service on Debian, Ubuntu, and other systemd-based distributions - no desktop environment, display, or GUI credential manager required. The Agent runs as a real non-root user, starts on boot, and restarts automatically if it ever stops; self-update is systemd-aware, so it verifies and rolls back through the same service instead of fighting its restart policy. Provider API keys and other credentials fall back to an encrypted local file when there's no OS keyring to use, which is the normal case on a bare server. See the Quick Start tab in this guide for the exact install command.

Remote operations require the Agent machine to be awake and connected. Background work can continue after you close the dashboard, but not after the machine shuts down.

Studio lets you point your own AI provider's coding agent (Claude, Codex, Gemini, Grok, DeepSeek, Qwen, GLM, Kimi, or Ollama) at a registered project directly from the dashboard, with a multi-tab source browser, live diffs, and a revert option for anything it writes. Rename or delete files from the tree, paste a screenshot straight into the chat for the model to see, attach text, Word, PDF, CSV, or Excel files for it to read, or start from a brand-new empty folder and have the agent scaffold a project from scratch. Opening an image file (PNG, JPEG, GIF, WebP, BMP, ICO, or SVG) shows an actual preview instead of a binary-file error - useful for checking something the agent just generated. The agent can also save what it learns as a reusable "skill" a later turn picks up automatically, kept either as real project files or in a machine-only folder that never touches the repo, your choice - view, download, add your own, or delete any of them from Configure Project's Skills tab, or see every project's at once from the dashboard's Skills Overview. On a shared machine, a team admin can scope access down to specific projects and choose exactly which AI providers a developer may use their key for, right from that same tab. Every write and command still runs locally on your machine, and provider API keys are stored only in that machine's own OS credential store - never in your account or on our servers.

It keeps going. An agent-mode turn stays running on your machine after you close the Studio panel or navigate away, and reopening it resumes the live view exactly where it left off - the step-by-step progress so far, and any pending command approval still waiting on your answer. The dashboard's Studio button also shows a live indicator whenever a project's agent needs your attention or is actively working, so you can tell at a glance without reopening it. It only stops if you click Stop yourself, or if the Agent machine itself goes offline.

Codex (ChatGPT) signs in with your ChatGPT account instead of an API key, picks its own model and reasoning effort, and shows how much of your plan's usage window is left. It plans and executes in one pass, so instead of an approve-the-plan gate you get a live read-only checklist of what it's doing. The Goal panel lets you set an objective and an optional token budget, then choose You drive (send each turn yourself) or Autonomous (the Agent sends the follow-up turns toward the goal on its own). An autonomous run stops at the budget, after 20 turns, when the goal is complete or blocked, or when you press Stop - and it pauses for your approval before anything outside the project or on the network. Like every agent turn, it keeps running on your machine even if you close Studio.