Index
diapason.workflow
¶
Workflow engine — DAG-based multi-agent pipelines.
WorkflowBuilder
¶
Fluent API for building workflow graphs.
Example
wf = (WorkflowBuilder("research_pipeline") .add_agent("researcher", agent="orchestrator", tools=["web_search"]) .add_agent("summarizer", agent="simple") .connect("researcher", "summarizer") .build())
sequential(*node_ids)
¶
Connect nodes in sequential order.
WorkflowEngine
¶
Execute DAG-based workflows.
Sequential nodes run in topological order. Parallel-eligible nodes (same execution stage, no inter-dependencies) run via ThreadPoolExecutor. Condition nodes evaluate expressions against prior step outputs. Loop nodes use LoopGuard from Phase 14.3.
run(graph, system=None, *, initial_input='', context=None)
¶
Execute a workflow graph end-to-end.
WorkflowGraph
¶
Directed acyclic graph of workflow nodes.
Supports DAG validation (cycle detection), topological sort, and execution_stages() for parallel-ready ordering.
execution_stages()
¶
Group nodes into parallel execution stages.
Nodes in the same stage have no dependencies on each other and can be executed concurrently.
topological_sort()
¶
Return node IDs in topological order (Kahn's algorithm).
validate()
¶
Validate the graph: check for cycles and orphan nodes.
load_workflow(path)
¶
Load a workflow definition from a TOML file.
Expected format:
[workflow]
name = "my_workflow"
[[workflow.nodes]]
id = "researcher"
type = "agent"
agent = "orchestrator"
tools = ["web_search"]
[[workflow.nodes]]
id = "summarizer"
type = "agent"
agent = "simple"
[[workflow.edges]]
source = "researcher"
target = "summarizer"