Skip to main content

Machines

A node is where your organization's tasks run. Each node holds worktrees: working copies of your projects. A host is the machine a node runs on; one host can hold several nodes. The Machines page lists your nodes, and on the Business plan its Groups tab holds your node groups; Hosts and Pools, its other tabs, are the operator's. Select a row to open its details, where its settings are changed. Cloud connections and NodeLinks are on Networks; the AI accounts your runners sign in as are under AI Accounts. Names are labels: two nodes may share one.

Open it from the sidebar: select Directory, then the Machines tab.

The node list​

The Nodes tab lists your nodes, one row each, whichever host it runs on, a page at a time. Under the list, Previous and Next turn the page, the line beside them says which rows you see, for example 1–50 of 1,204, and Per page shows 25, 50 or 100 rows. Select any column's heading to sort by it; select it again to reverse the order. The page, its size and the order are kept in the page's address too.

The Machines page: the search strip, the New node button and two online nodes

Above the rows, a strip narrows the list; every control is kept in the page's address, so a link to it opens the same view:

ControlWhat it does
SearchType and press Enter to find nodes whose name, tag or description contains the text, or whose identity key starts with it (four characters or more). As you type, it also suggests a Hosting (AWS, Google Cloud, Azure or On-prem) or an OS (Ubuntu, Amazon Linux, Linux, Windows, macOS, Docker or Distrobox); pick one to add it as a chip. Two chips of one kind show nodes that match either; chips of different kinds must all match. Select a chip's × to remove it.
Group byNone, Host, Hosting, OS or Node group. Each section is headed with its name and how many of this page's nodes it holds; nodes with none come last, under No host or Not reported, and nodes still being added under Preparing. Select a section's heading to fold it. Node group puts a node in two groups under both, and Not in a group last; it's shown where your plan includes node groups. The sort applies inside each section. The page remembers your pick.

On a phone, Group by sits behind the sliders button beside Search.

When nothing matches, the page says so and offers Clear filters.

Each node row shows, from left to right:

ItemMeaning
Round markThe node's status: a green dot when it's online. While a new node is on its way, a spinner marks it Deploying; a ! marks it Stalled when it hasn't connected after 15 minutes (longer for a machine whose usual boot is longer, such as Windows on bare metal) or its machine reported a fault. A ✕ marks a node its host couldn't start (failed). An online node whose machine hides other users' processes shows a ! instead of a green dot (its page says Can’t run work instead of Online): it runs no task until that is fixed. See Fix a node that can't run work. Hover the mark for why.
NameThe node's name. Select it to open its details. See Node details.
HostThe host the node runs on, when it shares a machine with other nodes.
LoadThree bars, top to bottom: CPU, memory and disk. A bar turns yellow, then red, as it fills. Hover them for the figures, for example CPU 12% · Memory 40% · Disk 55%, and the span they average over.
S/R/AThree numbers: slots / runners / accounts on the node, for example 1/0/1. Hover them for the words.
NV✓ when the node holds NodeVault keys.
NL✓ when one of the node's NodeLinks is mounted, ! when it has NodeLinks but none is mounted, blank when it has none.
CPUThe machine's cores, for example 8.
RAMThe machine's memory, for example 16 GB.
DiskThe machine's disk, for example 200 GB.
OSAn icon: Ubuntu, Amazon Linux, another Linux, Windows or macOS, or Docker or Distrobox when the node runs in that container. Hover it for the name, architecture and version, for example Ubuntu 26.04 · x86.
HostingAn icon for where the machine runs: AWS, Google Cloud, Azure or On-prem (your own machine). A node on a host shows its host's. Hover it for the name.
Last seenHow long ago the node last reported, for example now. Hover it for the exact time.
Enabled✓ (enabled), Draining or Disabled. Change it from the row menu or the node's details.
SelectA box to tick the row for a bulk action. Shown when your role has at least one.
⋯The row menu. See Row menu.

A node's NodeVault identity key is on its details, under Properties, then Identity; the search box also finds a node by the start of it. To check a node before trusting it with your NodeVault key, use the Fingerprint on Settings, then Features, then NodeVault, which your browser works out from the node's key itself. See How NodeVault works.

A node that paused itself because its disk is filling shows Paused: disk N% used (or Paused: disk full) under its row. On a node that shares its host's disk, hovering Load names that disk.

When the list is narrow, for example beside an open node or on a phone, columns give way so names stay readable: first Disk, RAM and CPU, then NL, NV, Host and S/R/A, then Hosting, OS and Enabled. Load goes last, so a phone still shows it.

While a node is still being added, for example on a host you just bought, the list shows a row named for the host with its step (Launching, Waiting for your machine, Connected, Joining or Placing), when it was last seen, and a line under it saying what it waits for, with how long: Install command not run yet… until the machine fetches its install script (as EC2 user-data, the command needs #!/bin/bash on its first line), then Installing…. While a machine of yours hasn't joined yet, the row's New install command gives org admins a fresh command on the New node page's progress, after confirming Get a new install command?; the one shown before stops working. It turns into the node's own row once the node is up, usually within a few minutes. A node whose host reported a fault shows that fault under its row. A failed add stays in the list with its reason under it; its ⋯ gives org admins Retry (for a machine launched in your cloud), Edit connection (when the cloud connection refused the launch) and Dismiss, which removes the row. Opening the link of a node whose cloud launch failed says didn’t start, with the reason and Back to Nodes.

A node your plan includes shows Included after its name, or Shared when it runs on a machine shared with other organizations. While Jaah prepares it, the list shows a row named Preparing and the product, for example Preparing Shared node · Small · 20 GB…. If it couldn't start, a warning above the list names the product and says what happened.

When your organization has no node yet, the page says No nodes yet.

Row menu​

Select ⋯ on a row to open its menu.

The row menu of an enabled Shared node: Drain, Disable and Delete

Which items appear depends on the node's state and your role.

ItemWhat it does
RetryShown on a node its host couldn't start, when it can be retried. Tries to start it again. Requires the org admin role.
UpdateShown when the node runs an older Jaah version than its organization's. Asks it to update; it applies on the node's own schedule. Unavailable, with the reason, while the node is offline or set not to update itself. Requires the org admin role.
DrainThe node takes no new tasks. It disables itself when the tasks it's still running finish. While a node drains, the menu shows Cancel drain instead. Requires the permission to edit nodes.
Disable / EnableStops or restarts new tasks on the node. Requires the permission to edit nodes.
Re-issueIssues the node a new token. The current token stops working at once, and the node stays offline until you run the new one on it. The new token is shown once. Requires the org admin role. Not offered on a Jaah Cloud node or one the operator allocated.
DeleteRemoves the node, in any state. See Delete a node. Requires the org admin role.

When a delete finishes, a short message at the bottom of the page says what was removed, and hides itself.

Bulk actions​

Tick the box at the end of a row, or the one in the column's heading to tick every row on the page, and a bar appears above the list saying how many are selected. Its actions run the row menu's verbs on every ticked node at once, and report each node that refused, in the server's words. An action that no ticked node can take is greyed with the reason.

ActionWhat it does
Enable, Disable, Drain, Cancel drainAs the row menu's. Disable asks first.
UpdateAsks every ticked node that runs an older version, and can update now, to update.
DeleteRemoves the ticked nodes, in any state, after you type their count. A node on a host is deleted alone; its host stays.

Groups​

Where your plan includes node groups (the Business plan), the Groups tab of the Machines page lists them, with All on top. Its columns, New group and each group's page are in Node groups; how to use groups is in Run a project on some nodes only.

New node​

Select New node to add a node, or New node on a host's row menu or page to add one to that host. Requires the org admin role. It opens the New node page beside the list, with four steps: Where, Configure, Name, and Create. On a phone, a line such as Step 1 of 4: Where stands in for the step bar. Next moves on, Back returns a step and Cancel goes back where you came from. Leaving any other way after a change, before the node is ordered, asks first: This node is not added yet. Save stays here to finish it; Discard drops the picks. Opened from a host, the page starts at Configure with that host picked. For the steps, see Add a node.

Where​

Headed Where will it run?, with a line under it: coding agents need a machine to run on, so rent one from Jaah or use your own cloud or computer. Pick one row; clicking one moves on to the next step.

The New node page on its Where step: the four steps, the Rent from Jaah row, which can't be picked, with Buy in Products under it, Launch in your cloud picked, Your machine, and Cancel and Next

ChoiceWhat it does
Rent from JaahWe run the machine, shared or dedicated. Not picked here: Buy in Products, under the row, opens Billing, where you add one under Add-ons › Add. Once it's set up, add more nodes to it with Add to a host.
Launch in your cloudA new machine in your AWS, GCP or Azure account, through one of your cloud connections. Your cloud bills you.
Your machineLaptop, desktop or server, in any cloud or office. Run one command on it.
Add to a hostOn a machine you already added. Free. Shown only when one of your hosts can take another node.

The way that suits your organization best is marked Recommended, and is picked when the page opens; when that way is Rent from Jaah, or can't be used, the first choice that can is picked, with no mark. A choice that can't be used is marked Unavailable and says why, for example No instance types are allowed. When Launch in your cloud is picked, its row adds Needs a cloud connection: a credential and a region. Pick it and the instance size next. When nothing can be added at all, a warning says why instead of the choices, and Next stays off.

Configure​

What this step asks depends on the choice:

ChoiceFields
Launch in your cloudConnection: the cloud account and region the machine launches in; Add new… makes one and brings you back. OS (Ubuntu or Windows Server, when your plan offers Windows), then Instance type: the machine's size, from the types allowed for the connection's cloud (Windows lists only the types that run it). A line under them says whose account it launches in and who bills you.
Your machineOS: Linux, macOS (Apple Silicon) or Windows (x64), those your plan includes. It picks the command you'll run, and can't change later. A line under it says what the machine needs first.
Add to a hostOne row per host, with where it runs and how many nodes it holds. A host that can't take another says why. With one host only, it's picked for you. Clicking a host moves on to the next step, unless the NodeLinks field follows.

The Configure step for Your machine: OS set to Linux, and Shape with Legacy picked over MicroVM

Below them, two more fields may show:

  • Shape: how the node runs its sessions: Legacy (sessions share the node), MicroVM (each session in its own VM) or Whole-node VM (the whole node in one VM). Shown only when more than Legacy is offered for this choice. MicroVM needs the MicroVM plan feature and a Linux machine. Whole-node VM needs the Whole-node VM plan feature, and is offered only on Launch in your cloud with Ubuntu or a Linux Your machine. MicroVM is picked when offered, except on Your machine, where Legacy is. A machine that turns out unable to run the shape picked gets Legacy. Change it later on the node's Shape tab.
  • NodeLinks: if your plan includes NodeLink, which of your organization's NodeLinks the node mounts. All are picked, so clear any it shouldn't mount; on Your machine none is, so pick those it should. Offered for your own cloud (only NodeLinks on its cloud), a node on a host and a Linux machine of yours.

Name​

Headed Name the node. A new node joins All, so every project on All can run its tasks there once it's online. To keep it for some projects only, add it to a node group: see Groups.

FieldMeaning
NameShown on the Nodes page and on tasks. Filled in for you on a node on a host; a name that can't be used says why as you type.

Create​

Headed Check and create. A summary of your picks: Where, then what you configured (for example Connection, OS, Instance type or Host), Shape when it was offered, Name, and NodeLinks when offered, then Create. A refusal Create can't get past by retrying keeps it off, with the reason, until you change a pick.

The Check and create step for Your machine: Where, OS, Shape Legacy, Name and NodeLinks None, with Create

Progress​

Once the node is ordered, the page is headed Adding and the node's name, and shows its Progress until it's ready; for Your machine, the command to run, with Copy command. Close leaves the page; the progress continues, and the list shows the node's row while it's being added. If adding it fails, the page says so and offers Retry; for a machine launched in your cloud, Edit connection also opens the cloud connection it used, where the fix usually is. A failure retrying can't fix offers Start again, back to the first step.

A new node installs the agents' command-line tools after it comes online. For up to an hour while it does, it takes no task that isn't already placed on it; such a task says Waiting: installing runner tools on and the node's name, and goes to another node if one is free.

Node details​

Select a node's name to open its details beside the list. The panel's top shows the node's name, its status, and its tabs: Overview, Properties, NodeLinks, Shape, GitHub Runners and Automations. NodeLinks appears only while your plan includes the NodeLink feature, and Shape only while it includes MicroVM; GitHub Runners appears while your plan's GitHub Runners limit is not 0. It also requires the org admin role and is not offered on a Jaah Cloud node or one the operator allocated. Automations requires the org admin role.

A Jaah Cloud node says A Jaah Cloud node: use it and place projects on it. Only the operator can change its settings. at the top of its details. A node paused by a filling disk shows a disk chip by its name. A node that can run no work says why in a warning above the tabs, for example This machine hides other users' processes, so the node can't watch its runners. Run it with its own process namespace, or on the host.

Overview​

The Overview tab of a Shared node: Last Seen, OS, Start on boot, Tailscale SSH, CLIs, Slots, Running tasks, Accounts, the Enabled, Drain and More buttons, then Metrics

When something is wrong, the tab opens with it: the node's active Findings, then one banner per issue. Hover an issue for its remedy. A healthy node shows nothing here.

A finding names the problem, what to do and when it was first and last seen: Memory critical, CPU critical, Process OOM-killed (naming the unit), a NodeLink store not answering, a stuck daemon step. Beside Memory critical, CPU critical, Process OOM-killed or a finding that session commands are timing out, Limit running tasks opens Properties at the node's Max running tasks, for those who may edit the node. A node whose machine never joined reads Never connected — the machine hasn’t joined yet., with the remedy to run the install command again and, for org admins, Install beside it.

FieldMeaning
Last SeenWhen the node last reported. When the node has stopped sending a report, it says Some reports are late; hover it for which report and when it last arrived. A node that has never reported shows Created beside it. When the alert above already says the node never connected, Last Seen is left out.
OSThe operating system and architecture (Arm or x86), for example Ubuntu 26.04 · x86. Hover it for the version.
Start on bootYes when the machine runs systemd, so the node starts by itself after a reboot; No, in a warning colour, otherwise. Hover No for why. Shown once the node has reported it. See What a node can't do.
Tailscale SSHYes when Tailscale SSH is on for the node; No, in a warning colour, otherwise. Hover No for why. Shown once the node has reported it.
RegionWhere the machine runs, by the region's name. Hover it for the region's code. Shown on a node in a cloud region.
CLIsThe agent command-line tools the node found installed, one chip each. A missing helper one of them needs is named in a warning colour. No agent CLI, in the same colour, means no task can run on the node; hover it for the fix.
SlotsHow many worktree slots the node holds. N retiring follows it to count slots on their way out.
Running tasksHow many tasks are running on it now.
HeldShown while tasks hold slots here: how many, and by whom, each task linked, for example 4 tasks waiting for your answer: #7745, #8609, #7744 +1 when all of them wait; otherwise asking marks each task waiting for your answer. Requires the permission to view worktrees.
AccountsHow many enabled AI accounts have a working sign-in on this node.

Below the fields are the node's buttons: Enabled, Drain, Re-enrol and More, which holds Restart and Delete. Hover one for what it does. On a node that has never connected, Drain and Re-enrol are greyed with Not connected yet.

  • Enabled: switches new tasks on or off for the node. Requires the permission to edit nodes.
  • Drain: the node takes no new tasks and disables itself when the running ones finish. Shown while the node is enabled. While it drains, the button reads Cancel drain. Requires the permission to edit nodes.
  • Re-enrol: re-issues the node's token, after you confirm Re-enrol this node?. The current token stops working at once; running the new command on the machine joins it again. Requires the org admin role. Not offered on a Jaah Cloud node or one the operator allocated.
  • Restart, under More: restarts the node's Jaah service after you confirm Restart node?. Running sessions keep going, and the node reconnects in under a minute. Greyed out, with Restart needs the node online, until the node is online. Requires the org admin role. See Restart a node.
  • Delete, under More: removes the node. See Delete a node. Requires the org admin role.

Below the buttons, Groups shows the All chip, then one chip per node group the node is in. Select a chip to open the group; × takes the node out of it, and Add to a group puts it in another. Changes apply at once; one that takes the node from projects with tasks running asks first. Shown where your plan includes node groups, and never on a Multi-Node Host. Changing it requires the permission to edit nodes.

On a node Jaah launched in your cloud, a section named for its cloud (for example AWS machine) shows the machine's Type, Disk and State (Running, Stopped, Starting… or Stopping…), with the buttons the machine offers now. Requires the org admin role. Where the machine can change type, Type is a field: pick another instance type and Save; the machine stops and starts again on it.

ButtonWhat it does
StopStops the machine after you confirm Stop machine?; the node goes offline until started. With tasks running, Force stop? lists them: their runs stop and they go back to Ready.
StartStarts a stopped machine.
RestartReboots a running machine after you confirm Restart machine?; its disk and type stay, and the node reconnects by itself. With tasks running, people's sessions open in its slots, or a node on it that stopped reporting, Restart anyway? says which: their runs end and each task's runner starts again in its slot. Offered on a node group's member too.
Grow diskSets a larger Size (GB). Disks only grow. With tasks running, Force grow? lists them, as for Stop.

Below them, each section shows only when it has something to show:

SectionMeaning
SessionsWhat runs on the node now, and on any node sharing its machine: each session's Task, Node, CPU, RAM and Processes, heaviest first.
MetricsThe node's resources over 1h, 6h, 24h or 7d. See below.
DiskThe node's disk cleanups and pauses, newest first.
Dead runsRuns on the node that died in the last 7 days, newest first: each run's task, why it ended and when. Pane log opens what its terminal last showed; a run with none says No pane log. Requires the permission to view sessions; Pane log requires the permission to edit them.
FindingsUnder Earlier (24h), problems the node found on itself in the last day that have since cleared.

Metrics shows the CPU, Memory and Disk bars, the figures (Memory used, Disk free, Disk total, Network in, Network out), and a chart for each the node reports: CPU, Memory, Runners (tasks running), Load, Pressure, Swap, Disk, /tmp inodes, Network in and Network out. The core count and total memory sit under their charts.

Between the figures and the charts, Signals shows the node's Load (its 1, 5 and 15 minute averages, beside its core count), Swap used, and CPU pressure, Memory pressure and IO pressure; Processes lists its busiest processes, each with its User, CPU and RAM. Either reads Not reported when the node doesn't send it. See Check a node's load, swap and busiest processes.

The Metrics card of a Shared node: the CPU, memory and disk bars, the disk figures, Signals and Processes, and the Runners chart over 6h

A node with no recorded history yet shows only its current figures and No history. Each chart shows the latest, lowest, highest and average value over the range. On the CPU, memory and disk charts, dashed lines mark 75% (yellow) and 90% (red). A hatched band marks a stretch with no samples, usually a node too starved to report; hover it for the times.

A node at 90% memory or more takes no new tasks or CI jobs until it has stayed under 90% for 10 minutes, and takes no more tasks than its memory holds (about 3 GB each). It also takes no new tasks while tasks have spent much of the last 10 minutes waiting on memory or disk, or while it runs as many tasks as its Max running tasks.

Troubleshooting, folded at the bottom, holds:

ItemMeaning
Reporting clocksHow long ago the node last sent each kind of report: Presence, Status report, Slot list and Running tasks list. A late one is marked STALE. Hover a name for what that report is, and an age for the exact time.
Config driftSettings whose value on the node differs from what was set for it, each read as declared → reported. Shown only when there are any.
Daemon configThe node's last reported setup, less what other tabs show. Requires the org admin role.

Properties​

The Properties tab: Name, Tags and Max running tasks, NodeVault and Identity key under Identity, then the Host fields

SectionFields
IdentityName, Tags, Max running tasks (empty: no limit; otherwise 1 to 64, and new tasks wait while that many run on the node; any other value keeps Save off. Hover it for a suggested number: one task per 8 GB of memory or per 2 processor cores, whichever runs out first), NodeVault (Holds keys or Can’t hold keys; hover it for when the node reported it and the fix), Identity key (the node's NodeVault key fingerprint, read-only, with Copy; shown once the node has one) and Description.
HostJaah home: the folder on the machine your projects are kept in. Auto-update: whether the node installs new Jaah versions itself: On, Off, or Default to follow your organization's setting, shown with that setting, for example Default (On); a node on a host follows its host. KVM for sessions (a node on a host you own): lets this node's sessions run KVM guests, such as the Android emulator; it asks first, because every node on that host then shares the risk.

Hover ⓘ beside a field for what it means. Change what you need, then select Save. Below the fields, Created by and Updated by (one Created and updated by line when it's the same person) say who made and last changed the node. Without the permission to edit nodes, the fields are read-only. Without the org admin role, only Name, Tags, Max running tasks and Description show, under Host properties are managed by an administrator.

The NodeLinks tab: No NodeLinks attached, and Shared build cache picked under NodeLink beside Attach

The tab lists the node's NodeLinks with each one's mount state, or No NodeLinks attached. Org admins pick one under NodeLink and select Attach, or Detach one after confirming Detach NodeLink?; either applies at once. With nothing left to attach, the tab says No other NodeLinks to attach.

Shape​

The Shape tab of a Legacy node: Shape Legacy, the Size from system, No report to show. and VM records None

SectionMeaning
ShapeHow the node runs its sessions: Legacy (sessions share the node), MicroVM (each session in its own VM) or Whole-node VM (the whole node in one VM), from those this node and its host can run. A change applies at once: Turn on MicroVM asks first, since each task then runs in its own VM; moving to or from Whole-node VM asks Change to …, as the node is installed again. When the node can't run MicroVM, the reason shows under it. Read-only where you can't change it.
VM sizeOn a Whole-node VM node: the vCPUs, RAM and Disk of its VM.
SizeOtherwise: the vCPUs, RAM and Disk each session's VM gets, and where each value comes from, for example from system. Not shown on a node that can't run MicroVM, unless MicroVM is already on.
ReportWhat the node last reported about its VMs: Hypervisor, Free memory, Free disk, Base image, Running VMs, Taps in use, Taps free, VMM processes, virtiofsd processes, VM folders and Reported, then the lists VMs on the node, Layers and Egress refusals. No report to show. until it reports.
VM recordsEvery VM Jaah has recorded for the node, with what it is for and its state; None when there are none.

GitHub Runners​

Sets up GitHub Actions runners on the node: you declare a pool and the node installs and runs it. With no pool yet, the tab says No GitHub Runners; select Set up GitHub Runners to start. Once the pool runs, each runner shows its state: Busy, Idle, Offline, Unregistered, Stopped or Gone, and a busy one names the workflow and job it is running. Under each runner: its memory limits and OOM policy, and its last five jobs with their results; Shutdown signal marks a job its runner was stopped under. The pool itself reads Running, Paused, Paused (disk), Paused: memory critical, Draining, Pausing, Applying or Not converging. While its runners are being removed, it reads Removing (Removal failing when the node keeps refusing), with Put back to keep the pool, Ask again to ask the node once more, and Drop record to forget the pool here; its runners then stay registered in GitHub.

Automations​

Requires the org admin role. Lists your organization's automations whose Target reaches this node: all nodes, this one picked, or all but some, this one not among them. It is the Automations list with the same columns, search and filters, narrowed to this node; the filters you set here are not kept in the page's address. Select a row to open the automation; closing it comes back here. New automation starts one with this node already its target. With none, the tab says No automations for this machine.

Delete a node​

Delete asks you to confirm first. Deleting a node:

  • removes it and every worktree on it, and stops whatever was using those worktrees: a running task or console session is ended, not asked;
  • asks the machine to erase those worktrees' folders;
  • terminates the machine if Jaah launched it.

Delete works on a node in any state:

  • Jaah Cloud node: its hosting ends, and unused days are credited or refunded, the same as Delete on its hosting.
  • Host that still has nodes: those nodes are deleted first.
  • Host still being added, for example one that says Waiting for you…: the install command stops working.
  • Unregistered node: it's removed from the list. It shows again if it reconnects.

When it is the last node on its host, the confirmation also offers Also delete host with the host's name, ticked, and says what that does: Its cloud instance is terminated., Jaah is uninstalled from the machine., or, for a Jaah-hosted machine, that the hosting ends and how unused time is returned. If the node goes but the host can't, the message names both.

This can't be undone.

Nodes the operator allocated​

The operator can give your organization one of its own machines. Such a node or host says Allocated by the operator at the top of its details. You use it like any other node: place your projects on it, and it runs their tasks. On an allocated host you can add and remove nodes.

You can still change its Name, Description and Tags, switch it on or off, drain it, attach NodeLinks and set its Auto-update. Everything else is the operator's: its other settings, its shape and its VM size are read-only, and Re-enrol is not offered. Delete still is.