Workflows: durable runs that fan out to agents
A workflow decides who works, in what order, what waits for what, and when a person checks. The Pyrana Harness keeps the run going until it finishes.
Unpack grows this view out of block 4 on the platform picture; Walk a run goes through the steps one click at a time and lights what each step uses; Fold shows it is all one configuration.
1How it starts
On requesta person starts it from a page, for the entities they choose
On a scheduledaily, weekly, any calendar
From another systemanother system calls it through the API
2What it accepts
Checked inputsa missing or malformed input is rejected before any work starts
Which entitiesan entity is whatever the work is for: a company, a store, a project
3One run
Starta person, a schedule or another system
Check inputsrejected here if wrong
Plan the workone task per agent and entity
Fan outagents run side by side
Wait and roll uplevels follow levels
Finishevery task done, results published
4How the work is shaped
Side by sideindependent agents work at the same time, for every entity
One after anotheran agent starts when the ones it needs are done
Up the hierarchyentities roll up level by level (for example store, region, the whole business)
A lead and its specialistsone agent hands focused questions to named specialists
5How it waits
For the whole levelthe next level starts when every entity below is done
Each parent for its owna parent starts as soon as its own entities are done
A person reviews between levelsthe run pauses until someone approves the next level
6Durable runs
Waiting is normala run can wait for data, a helper or a person, then continue
Automatic retriesa temporary failure in the run is tried again
Rerun one entityfix one failed entity without starting over
Who may start itonly people with access to the workflowgeneral only
7What it produces
Versioned results at every levelevery entity at every level publishes its own
The run's recordevery entity and agent run, with retries in its history
4Workflows
Hierarchieslevel by level, to the top
Durable runswait, resume, retry
Any workflow is one configuration
Startson request, on a schedule, or from another system
Acceptsthe inputs it checks and the entities it runs for
Agentswhich agents work at each level
Orderside by side, one after another, up the hierarchy
Waitsfor a whole level, or each parent for its own entities
Reviewswhere a person checks before the next level
Recoveryretries, waiting, rerunning one entity
Producesversioned results at every level
One workflow = one configuration.
Same shape, any process
Month-end close
Startson a schedule
Shapeup the hierarchy: entity, region, group
Reviewa controller approves between levels
Weekly store review
Startsevery Monday
Shapeside by side, one task per store
Producesa page per store for its manager
Contract intake
Startswhen another system sends a contract
Shapeone after another: read, check, summarise
Reviewlegal approves before it finishes
workflow.yaml Example from the demo, trimmed: workflows/demo_financial_analysis.yaml
workflow_key: demo.financial_analysis
mode: deterministic
orchestration:
type_order: [opco, group, holdco]
agents_by_type:
opco: [action_demo_sales_analyzer, action_demo_labor_analyzer,
action_demo_materials_analyzer, action_demo_pl_analyzer]
group: [action_demo_group_sales, action_demo_group_labor,
action_demo_group_materials, action_demo_group_pl]
holdco: [action_demo_holdco_pl]
wait_mode: per_parent # each group starts when its own companies are done
hierarchy_nodes:
- {id: PMW, type: opco, parent: IND}
# ... 8 more companies
- {id: IND, type: group, parent: PYRANA}
# ... 2 more groups
- {id: PYRANA, type: holdco}
agent_dependencies:
action_demo_pl_analyzer:
- action_demo_sales_analyzer
- action_demo_labor_analyzer
- action_demo_materials_analyzer
input_contract:
enforcement_mode: strict
json_schema: {required: [period]}