Nodes
A node is a machine your organization's tasks run on. It keeps worktrees, the working copies of your projects, and runs the runners that work in them. When a task starts, Jaah picks a node with a free worktree for its project and launches a runner there, so without a node that is online, enabled and open to the project, no task on that project can start.
A node can come from several places: bought from Jaah in Billing, launched in your own cloud account through a cloud connection, a machine you run yourself and register with a one-time command, or another node on a host your organization already has. Your plan can include nodes, and a node group can start cloud nodes for you; either way they appear on the Machines tab like any other.
A node shows two separate things. Its Status is ONLINE while it reports and OFFLINE once it stops. Whether it takes new tasks is Enabled, Draining (no new tasks; it disables itself when its running tasks finish) or Disabled. Neither draining nor disabling stops a task already running; Delete does, and removes every worktree on the node.
A node runs every project on All, or whose node groups include it, reaches the AI model through your model accounts, and fetches the credentials each task was given. The ideas behind all of this are in How nodes work.
- How nodes work — where a node comes from, its statuses and states, and how it relates to projects, slots, tasks and credentials.
- How many agents a node runs — what uses a node's memory and CPU, when to add a node, and mixing Shared, Dedicated and your own nodes.
- Add a node — launch a cloud machine, or register one you run yourself.
- Read a node's status — whether a node is online, how loaded it is, and whether anything is wrong.
- Find nodes that can't hold NodeVault keys — sort the Machines list by NV, check one node, and update the ones that can't.
- Drain or disable a node — stop new tasks on a node, with or without letting running ones finish.
- Drain or disable several nodes at once — tick the nodes, take them out of service in one step, and put them back.
- Run a project on some nodes only — every project runs on All; make node groups to keep a project on the nodes you pick.
- Change or delete a node group — rename a group, change its nodes, or delete it.
- Restart a node — restart a node's Jaah service, or reboot a machine Jaah launched in your cloud.
- Choose a node's shape — pick Legacy, MicroVM or Whole-node VM when you add a node, or change it later.
- Run GitHub runners on a node — set up a pool of self-hosted runners for a repository, then pause, resume or remove it.
- Recover runners whose removal is stuck — put a pool on Removing or Removal failing back, ask the node again, or drop the record and clean up in GitHub.
- Keep a node up to date — choose whether a node updates itself, and ask one or many nodes to update now.
- Delete a node — drain it, then delete it and, on its last node, its host.
- Name and tag a node — a name people recognize, tags and a note the Machines search finds.
- Change where a node keeps its project copies: point a node's Jaah home at another folder, such as a bigger disk.
- Limit how many tasks a node runs — set Max running tasks so a short-of-memory node stops taking more work than it can carry.
- Stop or start a node's cloud machine — stop a machine Jaah launched in your cloud while you don't need it, and start it again.
- Change a cloud node's machine type or disk — move a machine Jaah launched to another instance type, or grow its disk.
- Give a node a new token — replace a token that may have leaked or was lost, with Re-issue or Re-enrol, and run the new command.
- Change how big each session's VM is — set the vCPUs, RAM and disk each session's VM gets on one node, or inherit them.
- Set your organization's default VM size — set the size every session VM gets unless a project or node sets its own.
- Check what the virtual machines on a node are doing — read the Report on a MicroVM node's Shape tab: free room, live VMs, layers, refusals and VM records.
- Check whether a node is big enough — read a week of its Metrics, then resize it, cap it or add a node.
- Find what is overloading a node — read its findings, heaviest sessions, dead runs and disk, then limit or drain it.
- Find out why a node paused for its disk: read its Disk history: each pause, each cleanup and what it freed, then grow the disk or cap the node.
- Check a node's load, swap and busiest processes — read Signals and Processes on the Metrics card, where the node reports them.
- Fix a node that can't run work — a node marked Can’t run work because its machine hides other users' processes, and how to move it.
- What a node can't do — why a feature shows as off on a node's page, and what still runs there.
- Let a node's sessions run KVM guests — turn on KVM for sessions on a node of a host you own, for the Android emulator.
Every column, tab and field on the Machines page is described in the Machines reference. The worktrees a node holds are covered in Worktrees; folders nodes share, and the network they need, in NodeLink.