Get your free AI plan
Docs / Workflows and departments

Workflows and departments

What a workflow is, what is stored and what is derived, how a shared playbook behaves, who sees one, how departments apply, and the MCP fields to come.

What a workflow is

A workflow is a stored, named group of playbooks. It says which playbooks together deliver one outcome. A playbook can be in several workflows, in the same department or in different ones, and a workflow holds many playbooks, so the link between the two is many to many.

What is stored about a workflow is its name, its purpose, its department and its status. Nothing else is. Everything else a person reads about it (when it starts, how it is done, who is involved, what limits it and where it works) is derived from the playbooks, the rules, the policies and the map each time it is read, so it cannot drift from them.

The pages that list and open workflows arrive later, so the app does not show them yet. The playbooks themselves are in Workflows › Playbooks, and the sections below say what is true now and mark what arrives.

What is derived

The five parts of a workflow are read from typed links and ids, never from a name. A person sees each part as far as they may read it.

PartComes from
WhenThe rules that the workflow’s playbooks are linked to, and the playbooks that start by a trigger of their own: a signal kind they are bound to, or a schedule.
HowThe playbooks in the workflow.
WhoThe keys whose action policies cover the operations the playbooks use, and the people who approve: those for the workflow’s department and those for every department.
LimitsThe business policies that the playbooks and their rules link to, the policies scoped to the workflow’s department or to an operation the playbooks use, and the document rules for the playbooks or the department.
WhereThe objects the playbooks and their rules are drawn to, and the systems those objects live in or the playbooks work in.

Who, limits and where are shown but not required: a customer’s own agent can pick up a task without a key being named, and a workflow with no limit is allowed.

Status

A workflow has one of three statuses.

StatusStored or derivedWhat it means
RunningDerived from the partsIt has at least one playbook and something that starts it, and nobody paused it.
Needs setupDerived from the partsIt has no playbook, or nothing that starts it: no rule linked to its playbooks and no playbook with a trigger of its own. The status names the first part that is missing.
PausedStored: a person sets itA person paused it. It stays paused until a person changes that.

Only Paused is a person’s word. The other two follow the parts each time, so a workflow whose rule is removed shows Needs setup without anyone changing it, and one that gets its rule shows Running.

Beside the status the workflow pages show how many of its tasks are Waiting on a person. That is a count of the workflow’s tasks, taken from the task statuses of Work › List. It is never a fourth status.

A playbook, rule or policy in more than one workflow

A playbook, a rule or a policy that is used by more than one workflow is stored once. It is not copied into each workflow, so a change to it takes effect in every workflow that uses it.

The workflow pages mark a part used by more than one workflow Shared, with the workflows it is used in.

A playbook that is shared has a page of its own, opened from its title in Workflows › Playbooks. The page says "This playbook is shared", names the workflows a change reaches, and offers Make a copy for one workflow: a new playbook with the same text, anchors, trigger rules and limits, and the links to what it works on and in, in use for the workflow you choose, which stops using the shared one while the other workflows keep it. The copy starts with no schedule: a schedule stays with the playbook it was set on. It starts on the same kinds of signal as the shared playbook, so both start on the same signal until you change one. A rule or a policy that is shared has no such page yet, so a change to one is not warned about.

Arrives with a later release. A rule, or any other kind of alarm, that no workflow handles is marked “Nothing handles this yet: no workflow”, with the action Make a workflow; a playbook that handles a kind outside any workflow counts as handled and is listed as not in a workflow yet. None of this is in the app yet.

Tasks

No task records which workflow it came from. The workflows of a task are derived each time. A task that a rule raised belongs to the workflows that hold a playbook linked to that rule. Any other task belongs to every workflow that holds a playbook that is guidance for it: by its kind, its objects, its suggested operation, or the playbook that a routine or a thread names. A task that no workflow handles has none. This says which workflows own a task, not which playbook an agent opened. A signal over MCP is a task in the app (see Words).

A rule that is in several workflows is meant to raise one task per workflow, so that one workflow closing its task never closes another’s. That change to how rules raise signals is a named follow-up, Rule signals per workflow, and it has not shipped. Today a rule in two workflows raises one task, and that one task belongs to both workflows.

Who sees a workflow

A key or a member limited to part of the map sees a workflow only if it has at least one playbook and every one of its playbooks is in that scope. A playbook that is bound to nothing, or only to signal kinds or operations, is outside a limited scope; a bound object or process must be in the grant, as the map stands now. A key or member whose scope is the whole map sees every workflow, one with no playbook yet included. Key shapes says what a limited key sees and does.

Departments

A workflow’s own department is the stored one. It can be empty: a workflow with no department yet shows as having none, and Maplith never guesses a department from its playbooks.

A playbook carries no department of its own. It shows under one by the first of these that applies:

SourceWhenThe playbook shows
The workflow’s departmentThe playbook is in a workflow that has a department.Under the department of each of those workflows: a playbook in workflows of two departments shows under both.
DerivedThe playbook is in no workflow that has a department, but it has an owner or systems that belong to a department.Under the departments derived from its owner, or from its systems.
NoneNeither of the above.Under Unassigned.

A department admin writes only the workflows of their own department. Departments are managed in Settings › Workspace › People › Departments, where they are switched on for a workspace.

Over MCP

These tools return a workflow and a department, as the table says.

ToolWhat it gainsState
get_task_contextworkflow (id, name, department) and department for the task; workflows lists every one when there are several.Served
list_signalsThe filters workflow and department; each signal carries its workflow (and workflows when there are several).Served
searchFinds workflows; the result id is workflow:<id>.Served
fetchReads workflow:<id>: name, purpose, department, status, playbooks, and the derived rules, policies and objects.Served
get_objectused_by_workflows: the workflows whose Where names the object.Served
get_map_overviewdepartments, each with its counts of workflows, playbooks, rules and policies.Served
get_value_summaryOptional group_by workflow or department: signals and their records per group, the rest under no_workflow; findings and metrics stay ungrouped.Served
draft_playbookAn optional workflow with part whole, to place the new draft in one; the answer says workflow_placed.Served

A key limited to part of the map sees a workflow by the rule above and only the departments of the workflows it sees, and is never told the name or a count of one it may not see. Rules and policies are named and counted for a key that sees the whole map only. Who acts and who approves is not served: key and person identities are not a workflow’s content for an agent. Until Rule signals per workflow ships, a rule in two workflows has one signal, placed in both.

Where to go next

  • How Maplith works: the red thread in five steps.
  • Words: what each name means, and what it was called before.
  • Key shapes: what a key may see and do, and a key limited to part of the map.
  • Work the loop: an agent’s main path, from a task to a record of the work.
  • The MCP reference: every tool, with its inputs, an example and its refusals.
  • Migration: which tools are being merged, in which order and until when.
  • Signals and rules: how a rule raises a signal.
  • Agent keys: what a key sees and may do, and its policies.

← All docs · Markdown version