Your Agents Can Operate DorkOS
Ask an agent to create agents, schedule tasks, install packages, change settings, read your activity, and explain how DorkOS works, in plain language
Your Agents Can Operate DorkOS
Your agents don't just write code. They can run DorkOS for you. Ask in plain language, and an agent can set up new agents or put work on a schedule. It can add tools from the marketplace, change your settings, and tell you what's been happening. It can also explain how any of it works.
This is a slice of the app, not all of it. Agents reach the parts listed below. They cannot open a session for you, set up a workspace, connect a runtime, or run git. Those stay yours to do by hand.
You stay in charge of what lands. Anything an agent cannot take back puts a card in front of you first: removing an installed package, deleting a scheduled task, removing an agent, and changing an agent's safety boundaries. Adding a package or creating one does too, because both bring new code onto your machine. Smaller edits, like changing a preference or renaming an agent, go through without a card, and they land in your activity feed after the fact. When DorkOS knows which agent asked, the feed says so by name. The settings that protect your instance are a third case: agents are refused those outright, with no card to say yes on.
One thing to know about who gets to answer those cards. With Require login off, which is how DorkOS starts, the agent that asked is refused: it carries an identity, and so does the retry token it was handed, and either one is enough for DorkOS to turn it down. What DorkOS cannot do without a login is tell you apart from a program on your machine that deliberately hides both. Turn Require login on and only your signed-in account can answer. Who gets to answer has the whole picture.
What you can ask for
Create an agent
"Set up a new agent for my side project in ~/code/blog." The agent creates the workspace and registers it, ready to use.
Schedule a task
"Every morning at 8, have an agent check my open pull requests." It creates the scheduled task for you.
Add tools from the marketplace
"Find a code-review pack and install it." The agent searches the marketplace, shows you what it found, and installs it after you confirm.
Change a setting
"Keep the git branch in my status bar." The agent pins it, so it shows even when there's nothing to report — until the window gets too narrow to hold it.
Read your activity
"What have my agents done in the last hour?" The agent reads your activity feed and summarizes it.
Check for updates
"Am I on the latest DorkOS?" The agent checks your version against the newest release.
Ask how DorkOS works
"How does Relay work?" The agent searches these docs, reads the page that answers, and tells you what it says instead of guessing.
New agents already know how
Every new agent you create comes with a set of built-in skills that teach it how to drive DorkOS: how to create agents, schedule tasks, use the marketplace, change settings, and read activity. Another one teaches it to look things up in these docs, so a question about DorkOS gets the documented answer rather than a guess. DorkBot, your built-in assistant, gets them too. So you can just ask; you don't have to explain how DorkOS works first.
A few things to know
- Your agent is happy to edit itself. It can adjust its own name, personality, and notes. It's told to check with you before editing a different agent, which is instruction, not a lock: DorkOS does not stop it. Watch your activity feed if you have agents you don't want touched.
- Some agents are protected. Built-in agents like DorkBot keep their identity, and this one is enforced: an agent can't rename or impersonate them, however it asks.
- Most settings changes are not gated, but the ones that protect your instance are refused. An agent can change a preference without a card, and the change shows up in your activity feed. It is simply refused the settings that decide who can reach DorkOS, what it can reach, where your keys go, what leaves the machine, whether you get asked before an agent acts, how much your agents may say and spend on their own, and any limit you turned down to give your machine room. There is no card for those and no way to say yes, because they are yours to set. See Settings your agents cannot change.
- Anything you cannot take back waits for you. Removing an installed package, deleting a scheduled task, removing an agent, and changing an agent's safety boundaries each put a card in front of you. So do installing a package and creating a new one, because both bring new code onto your machine. Nothing lands until you answer. See Action Approvals.
Doing it from the command line
Codex and OpenCode agents don't get DorkOS's in-app tools by default, so the dorkos command line is how they reach DorkOS. (An experiment, DorkOS tools in every runtime in Settings, hands them the same in-app tools your Claude Code agents get — off until you turn it on.) The command line works the same for you in a terminal. These commands talk to your running DorkOS, and every one takes --json for clean, scriptable output.
dorkos capabilities # Ask DorkOS what it can be told to do
dorkos call operator.check_update # Run one of those by name
dorkos agent list # See your agents
dorkos task list # See your scheduled tasks
dorkos task trigger <id> # Run a task now
dorkos activity --limit 20 # Read recent activity
dorkos version --check # Compare your version to the latestdorkos capabilities is the useful one for an agent: it asks your running DorkOS for a current list of the actions it accepts by name, so the agent doesn't have to guess from docs that may be out of date. dorkos call then runs any of them. That list is not every tool DorkOS has, and it says so, so an agent knows to look at its own tools for the rest.
If you turned login on, give the CLI an API key first, or these commands will tell you they need one: see If your DorkOS asks you to sign in.
See the CLI reference for every command and flag.