Projects and workspaces
Can a project have no repository?
Yes. Clear Repo for work outside a codebase, such as research or writing. The project then has no repositories, branch policy, gate or stack commands. A task starts in an empty slot on one of the project's nodes, and its outcome is its report rather than a merged pull request. Ticking Repo again brings the old values back. Only an org admin who can create and delete projects can change Repo on a saved project.
See How projects work and Add a project.
Can a project's tasks run on more than one node?
Yes. A project runs on All, every enabled node of your organization, or on the nodes of its node groups, all at once. Jaah chooses a node for each task, preferring the least busy, so tasks of one project can run side by side on different nodes. A task stays on the node it started on until its slot is let go, for example when its runs keep failing to start there or you reset it. See Run a project on some nodes only.
See How worktrees work and How nodes work.
Does stopping a task give up its slot?
No. A slot stays with its task until the task is completed, deleted or reset, an operator releases it, or one of the cases under Which node a task runs on lets it go. So a task you stop and start again carries on with the work already in its slot. To see which tasks hold a node's slots, open the node's details and look under Held.
See How worktrees work and Give a project more slots.
Can I try a task's change before it merges?
Yes, when the project runs a stack in each slot: each task holding a slot has its own copy of the app running with its change. On the board, select ⋯ on the task's row, then Stack; or on the task's Details tab, select 🌐 slot ↗ beside Runner. The app opens in a new tab, on the task's own branch, ports and data. The link shows only while the task is in progress and its slot's stack is up, and only to people who can view worktrees, as org admins can. Once the task is done, reset or deleted, the link goes.
See Try a task's change in its slot and Run a project's stack in each slot.
What's the difference between Start and Customize on a template?
Start asks for the Project, and for any inputs the template needs, then starts the task at once and opens it. Customize opens the new-task form already filled in from the template, so you can change the brief, persona or any other field before you start it. Either way, the task's page shows From template.
See How templates work and Start a task from a template.
What's the difference between a persona and a playbook?
A persona is the kind of expert a runner acts as, such as Coder or Researcher. It decides how the runner works and what it hands back. Every task runs as one persona. A playbook chains experts together: an ordered list of stages, each pairing an Expert with the Deliverable it hands back.
See How personas and playbooks work and Choose a persona for a task.
Can I choose which model account my task runs on?
No. Each time a task starts a run, Jaah picks an enabled account that has room and fits the task. Subscriptions are picked first; an API-key account runs a task only when no subscription can take it. You can steer it with the task's Provider: Any runs on accounts with Auto-assign on, and a named provider runs only on that provider's accounts.
See How model accounts work and Add a model account.
What happens when a model account runs out of quota?
A subscription whose quota is used up, or nearly so, is skipped until its quota resets: its row shows ▲ and ↻ shows when the quota resets. New runs go to another account that fits the task, and an API-key account takes a task only when no subscription can. If the account runs out during a run, that run fails and uses one attempt; the task is queued again and its next run picks from the accounts that still qualify. To keep work moving, add a second account.
See How model accounts work and Add a model account.
What's the difference between a custom agent and a registry agent?
Both are agents that run a task's work instead of a built-in runner, and both are picked with the task's Agent field or a project's Default agent. A custom agent is one you build: you register its GitHub repository with Add agent on the Custom agents tab, and Jaah follows its branch or tag. A registry agent is a ready-made coding agent from the public ACP registry: an org admin selects Install on the Registry tab, it stays at the version installed, and one that needs a key uses one of your API-key model accounts. Both are listed on the Custom agents tab once added.
See How agents work and Install an agent from the ACP registry.
Can we write our own templates?
Yes, with the Templates (org) feature on your plan and permission to edit templates. People with both see Manage and New template on the gallery, and Duplicate on Official rows, which copies the template into your organization for you to change. Your own templates show Your org on their rows.
See How templates work and Create a template.
Where are Templates, Playbooks and Agents in the sidebar now?
Under Settings, in its Customize group: Playbooks, Templates, Agents, Compliance Packs and Pipelines, and Agents holds the Custom agents, Catalog and Registry tabs. You see only the items your plan and role allow. The model accounts runners sign in as are under Directory, then AI Accounts.
See How templates work, How personas and playbooks work, How agents work and How model accounts work.
What happens to my task when its node goes offline?
It waits for that node. A task keeps its slot, so it comes back to the same node with its work. If the node won't be back soon, open the task, select ⋯, then Reset, and confirm. That lets the slot go; select Start and the task takes a slot on another of the project's nodes. A reset discards what the task had not pushed yet. If the task is pinned to the offline node, set Node to Any first.
See Move a task off an offline node and How worktrees work.
Can a template run on a schedule?
Yes, through an automation. Select Audit, then Automations, then New automation, and set the Schedule, such as every Monday at 09:00. On What, pick the projects, select Template instead of Body, pick the template and fill in its inputs. Each run files one task per project from the template as it is at that moment, so editing the template changes the next run. To check each task before it runs, set Start to Leave as draft on the automation's Settings tab. Only Org Owners and Admins see Automations.
See Run a template on a schedule and How templates work.
Can I stop a playbook's stages asking me questions?
Yes, two settings decide it. Full Auto on the playbook's Playbook tab is the default: On means stages run without stopping to ask. Each stage's Asks questions then picks When Full Auto is off or Never asks, so an analysis stage can ask while the build stages never stop. Open Settings, then Playbooks under Customize, select the playbook's ⋯, then Edit, and select Save when done. Built-in playbooks are read-only: duplicate one first. Changing a playbook is for Org Owners and Admins.
See Choose when a stage asks questions and How personas and playbooks work.
How do I run more of a project's tasks at once?
Add another node. A project on All runs on it as soon as it's online; a project on node groups needs the node in one of its groups. Each running task holds one slot, so when no node has room, new tasks wait as Queued.
See Give a project more slots and How worktrees work.
How do I make every task in a project use the same agent?
Set the project's Default agent. In the sidebar, select Directory, then the Projects tab and the project's name. In Settings › General, pick the agent in Default agent, then select Save. Every task whose Agent is left on Default runs on it; a task that picks another agent overrides it. Any leaves it unset, so tasks run on the first free agent. Anyone who can edit the project can set it.
See Set a project's default agent and How agents work.
Why did my task start again on another node?
A task keeps its slot, and so its node, until the slot is let go. That happens when its runs fail to start twice on that node within a day, when you reset it or move it to another project, when the node is removed from the project or deleted, or when it waited on your answer until Jaah stopped it with nothing unsaved in its slot. Its next run then starts in a new slot, on whichever of the project's nodes is least busy. It begins from what it last pushed: anything it hadn't pushed was lost with the old slot.
See How worktrees work and Move a task off an offline node.
Why is an agent greyed out in a task's Agent field?
It can't run this task, and its reason follows its name. Most often the agent doesn't serve the task's persona: pick another persona, or have the agent's owner add this persona to it. Other reasons: it is disabled, it was taken out of the catalog, or it is marked Unsupported until its owner registers it again.
See How agents work and Fix a custom agent whose refresh failed.
What's the difference between a basic and a custom pipeline?
A basic pipeline is ready-made, and every plan can pick one in Settings › CI Pipeline: PR into integration branch, where merging a pull request deploys to the live stage, or Integration + Release, where merging deploys to Dev and a release moves it to Production. Its stages, gates and branch prefix are set for you. A custom pipeline has your own stages, steps, approvals and rules, edited in the project's Settings › CI Pipeline. Custom pipelines are a plan feature.
See Choose a basic pipeline and Set up a project's pipeline.
How can I see which tasks used a slot before?
Find the slot on the task's Details tab: Runner names the node and the slot, such as · slot 1.
Then select Audit in the sidebar and the Sessions tab. Set Window to All time, set Node
to that node, and type the project's name, a hyphen and the slot number in Search, such as
website-1. Task names the task each run was for. Viewing Sessions needs a role that can view
sessions, and your plan must include Session transcripts.
See See what ran in a slot and How worktrees work.
Why does a node say some of its slots are retiring?
Jaah makes a slot for each task as it starts and removes it once the task lets it go. While the node deletes a slot, the node's details show it after Slots, such as 2 · 1 retiring. A retiring slot is no longer counted as one of the node's slots and no task can take it. It disappears once the node has deleted it. Anything a task hadn't pushed from that slot goes with it.
See How worktrees work and Free a held slot.
Can a template ask for details each time it starts?
Yes. In the template's Description, type {{input: Label}} where the value belongs, for example
Fix this bug: {{input: The bug}}, and select Save. Start then shows one field per input and puts
each answer in the brief. Add a question mark, {{input?: Label}}, for one people may leave out. Only
Start and an automation that files tasks from the template ask; Customize leaves the marks in the
brief for you to replace. Editing templates needs the Templates (org) plan feature and permission to
edit templates.
See Ask for details when a template starts and How templates work.
Can a playbook build each item of a plan on its own?
Yes. Give the playbook a stage whose Deliverable is Plan, with Planner as its expert, then a stage that the plan creates and that Runs Once for each planned item. When the playbook runs, the plan files one task per item and the repeating stage runs for each. Add a stage after it, such as a review, to collect the items: the repeating stage can't be last. The built-in Epic playbook works this way. Editing playbooks needs the Owner or Admin role.
See Repeat a stage for each planned item and How personas and playbooks work.
Why isn't my latest change to a custom agent running?
Jaah re-checks a custom agent's branch or tag after each push and pins its newest commit. When that check fails, the agent is marked Last refresh failed and keeps running the last commit it pinned. Open Settings, then Agents under Customize, and the Custom agents tab, and select the agent: its warning gives the reason. Fix it, then select More, then Refresh. Active means the new commit is pinned. An agent marked Never fetched runs nothing until its first check passes.
See Fix a custom agent whose refresh failed and How agents work.
Can a Live Cast show a customer's name?
A built-in privacy check stops it. Whatever your default and rules say, it hides any task whose title, description or outcome names a customer, another organization you work with or a person, or carries personal details such as an email address or phone number. Vendors, tools and your own project and organization names don't count. It doesn't read comments or artifacts, so check those yourself and set the task to Hidden if they need it. If the project's Live Cast settings say Tasks aren't being checked., nothing is checked yet: ask an operator to turn task checks on.
See Publish a Live Cast and How projects work.
Why can't I press Release?
Hover the greyed Release button: it says why. Nothing to release means no merged change is waiting for the next stage. Waiting for tests means changes are waiting for the merge lane to test them; someone allowed to ship untested changes can still release them. You can't release to and the stage's name means that step isn't one you can release from here. A step that is closed, for example until its last arrival is verified, shows its own reason.
See Release to production and Release changes that have not been tested.