Aether is a self-hosted runtime for AI coding agents. You run the server on a Linux machine you own, each run gets its own container and git worktree on it, and the work comes back as a git branch. The quickstart walks through the whole loop.
Yes. You install and run the server yourself, and it is the only thing the dashboard and the phone app talk to. There is no hosted service in the middle, and the CLI reaches out only to download its own updates. The privacy policy states that nothing about you reaches the people who publish it. Read the privacy guide.
A hosted platform runs the agent on infrastructure the vendor operates. Aether runs it on hardware you operate, with your own vendor logins, and hands you a branch to review. There is no account and no per-seat fee, because there is nothing for anyone to bill you for. The comparison page puts the two side by side.
No. There is no signup and no password. The first identity to link a fresh server becomes its admin, and everyone who links after that joins as a collaborator, pending an admin's approval unless they came in on an invite code. See the quickstart.
Yes. Aether is published by 3xDevOps under the GNU General Public License version 3, and that covers the server, the CLI, the dashboard and the Android app.
On GitHub, at github.com/3xDevOps/aether. The server and the CLI are written in Go, and the dashboard is a static web bundle embedded in those binaries.
Yes. Claude Code is one of the shipped harnesses. Install the CLI once in your environment terminal, log in with its own /login flow, then start work with aether run "..." --agent claude. The run happens in a container on the server, not on your laptop. See the harnesses guide.
Yes. codex is a shipped harness, launched with --agent codex. You sign in with the Codex CLI's own device-code flow inside the environment terminal, or an admin sets OPENAI_API_KEY in the server's environment instead. See the harnesses guide.
Yes. opencode is a shipped harness. You log in with opencode auth login inside the environment terminal, and the credentials stay in your member home. It is launchable for runs, though it is not one of the agents the dashboard's guided setup step covers. See the harnesses guide.
Seven harnesses ship with it: claude for Claude Code, codex for the OpenAI Codex CLI, pi, omp for oh-my-pi, opencode, fake for testing, and custom for a definition the deployment supplies. A member can register another CLI with aether agent add, and an admin can pin definitions server-wide. The harnesses guide lists what each one needs.
No. Aether does not install agent CLIs. You install the one you want in your own environment terminal, into ~/.local/bin. For Claude Code, Codex and pi the dashboard opens that terminal and types the vendor's install command, but the install and the login are still yours to complete. See the harnesses guide.
You sign in through each vendor's own flow inside the environment terminal. The credentials land in your member home on the server and are mounted into every container you get, so all your runs share one login. An admin can set ANTHROPIC_API_KEY or OPENAI_API_KEY in the server environment instead. Aether never copies credential files back to clients. See the security posture.
Yes. A run is a container on the server, supervised by the server, and its terminal session lives there too. Detaching or closing your laptop does not stop it; reconnect later with aether attach <run-id> or from the dashboard. See how a run works.
Each run is one container with its own git worktree and branch, started from your saved environment image, or from the server's standard Ubuntu image if you have not saved one. The run's checkout is mounted at /workspace and your persistent member home is mounted as $HOME. See the environments guide.
Yes. Docker is required on the server host, and every environment terminal and run is a container on it. The runtime layer is written as a seam so other backends could be added later, but Docker is the only one implemented today. See the install guide's requirements.
Aether commits the work and publishes it to a branch named aether/run-<slug>-<short-id> in the server's repo, then aether pull <run-id> brings that branch into your clone without switching to it. A headless run destroys its container at that point; a run with a terminal UI stays up until you close it. See how a run works.
No. Aether never switches branches and never merges a run into your base branch. You review the branch and merge it yourself. If you connect GitHub, an agent can push its own branch and open a pull request as you, but the merge is still a decision you make. See the quickstart.
Yes. One server hosts everyone's agents side by side, and a solo setup is just a team of one. There are three roles: viewer, collaborator and admin. The first person to link the server becomes the admin, and everyone else joins as a collaborator, pending approval unless they used an invite code. Collaborators can launch runs, steer them and kill them. See what a team looks like.
Yes. Anyone who can view a run gets a live read-only mirror of its terminal, and several people can attach at once. A collaborator can type into a run, inject an instruction, pause it or kill it, and every act is attributed in the run's timeline. aether protect limits steering and killing one run to its owner and the admins. See the teams guide.
Yes. An admin sets a per-workspace budget in dollars, with a hard cap and a softer warning below it. At the cap, new runs are refused and the runs already going finish. Runs whose harness reports no token usage count as unmetered, so the totals are a floor rather than the real spend. See the teams guide.
The server is Linux only, on amd64 and arm64. The CLI is a supported client on Linux, macOS and Windows. There is an optional desktop app you build locally on any of the three, and every release carries an Android APK for the dashboard. See the install guide.
No. macOS and Windows are client platforms: you run the CLI there and point it at a Linux server. The install script refuses to put the server on anything else. See the install guide.
That is the intended setup: a home server, a spare PC, or an old laptop repurposed to stay on all day. The server needs Linux on amd64 or arm64, plus Docker and git. No minimum CPU or memory is documented, so size the machine against the work you expect to give it. See the install guide's requirements.
Yes, and it is the recommended setup. Put the server on your tailnet and aether link <server> identifies you through Tailscale, so there are no SSH keys to hand around. Set a web port and the server can host the dashboard itself over HTTPS on its tailnet address. Invite codes cover anyone outside the tailnet. See the networking guide.
On Linux and macOS, run the install script from the repo and choose server or client. A server then needs sudo aether-server setup and the systemd unit enabled. Windows has a PowerShell script. You can also download the release binaries by hand and check them against checksums.txt. See the install guide.
No. Your repository lives as a bare git repo in the server's data directory, on the machine you own, and you push to it over the server's SSH port. The people who publish Aether run no service, hold no copy, and cannot delete anything on your behalf. See the security posture.
Aether runs the containers on your server, but the agent CLIs inside them talk to their own model providers using your logins, so your prompts and the code they read reach whichever provider you chose. Beyond that, the server fetches releases from GitHub, pulls its standard image from ghcr.io, and reads vendor usage endpoints to show your quota. Nothing is sent to the people who publish Aether. See the privacy guide.
