Plug-ins in 📝MythOS are programs you install on one of your 📝Devices so it can take loop runs that need more than a plain agent session, such as a run that should open a coding workspace, follow a review checklist, or drive a tool of your own before the agent starts. Most loops never need one, but some work has a shape of its own, and a plug-in lets you teach a device that shape once and send it every run that needs it. They are for 📝Oracle members running Loops on their own machines, and they open together with Loops.
Key Capabilities
- Lives on your device — a plug-in is installed and stored only on the machine; MythOS never sees its code
- Routes runs — a loop that requires a plug-in goes only to a device that has it installed
- Shown by name — the device page lists each device's plug-in names, never a command or file path
- Keeps the Session — MythOS still records the run's Session, transcript, asks and replies around the plug-in
- Starts your agent safely — the plug-in launches the run's agent through a command the runner has already vetted
- Finishes with real output — a plug-in completes a run with a memo, a pull request or a channel message
Getting Started
- Set up the device first; see 📝Devices.
- On that machine, install the plug-in:
mythos-device plugin add <name> <command…>. The name is the capability a loop will ask for, and the command is the program to run, for examplemythos-device plugin add code-review /usr/local/bin/review-handler. Add--env NAMEfor each variable from the runner's environment the program needs. - Check it with
mythos-device plugin list. Within one heartbeat, the plug-in's name appears under Plug-ins on the device's page in Loops → Devices. - In the loop's Policy, set
Runner capabilitiesto that name. From then on, its runs go only to a device that has the plug-in. - To take it off a device, run
mythos-device plugin remove <name>.
FAQ
- What does MythOS store about a plug-in? Only its name, which the device reports with each heartbeat and each request for work. The command, its path and its code stay in the device's own runner config, and no key, MCP tool or loop can install, change or read one.
- Why list it on the device page at all? So you can see which machines can take which loops. When a loop's runs sit waiting with "no capable runner", it means no online device has the plug-in that loop requires, and the device page is where you check.
- How does a plug-in talk to MythOS? Through files in the run's folder on the device. It reads the run's payload, posts asks and reads the answers, reports progress, starts the run's agent, and writes the run's outcome and output. The runner relays all of it to the run's Session.
- Can a plug-in run hold for my approval or run script steps? Not yet. A plug-in run takes only gates set to approve automatically. A loop that needs a person's approval mid-run, or script steps, is refused before it starts on a plug-in.
- Is a plug-in safe to install? It runs as you, on your machine, with the run's limited environment plus only the variables you name. Install only programs you trust, exactly as you would any command-line tool.