API reference¶

The Python SDK exposes two entry points: ControlPlaneClient for managing/observing workflows, and BoundFlowWorker for running them. This page is a high-level map; see the docstrings in sdk/python/boundflow for full signatures.

Warning

Pre-1.0. The SDK surface and the underlying gRPC protobufs may change before 1.0.

`ControlPlaneClient`¶

Construct with a server address and API key (both default from the environment):

async with ControlPlaneClient(api_key=...) as cp:
    ...

Tenants & tenant groups¶

Method	Purpose
`create_tenant(name)`	Create a tenant in the caller's tenant group.
`list_tenants()`	List the caller's tenants.
`create_tenant_group(name)`	Create a tenant group.

Workflows¶

Method	Purpose
`create_workflow(type, tenant_id, config=…)`	Register a workflow.
`activate_workflow(id)`	Move a workflow to `active`.
`invoke_workflow(id, …)`	Trigger a run; returns a `request_id`.
`delete_workflow(id)`	Delete a workflow.
`list_workflows()`	Every workflow with its lifecycle / workflow state.
`get_workflow_lifecycle_state(id)`	Current lifecycle state.
`get_workflow_state(id)`	Current workflow (enablement) state.

Runs¶

Method	Purpose
`list_workflow_runs(id)`	Run history with per-run outcomes.
`get_request_info(request_id)`	Status + outcome of a single run.
`resolve_interrupted_workflow(id, request_id)`	Clear an interruption and re-activate.

Policies & pricing¶

Method	Purpose
`set_agent_runtime_policy(id, agent, policy)`	Hard per-run caps.
`set_agent_lifecycle_policy(id, agent, rules)`	Post-run model switching.
`set_workflow_lifecycle_policy(id, rules)`	Cooldown / rollback / pause.
`set_model_pricing(model_id, …)` / `list_model_pricing()`	Per-tenant-group pricing.

Approvals & audit¶

Method	Purpose
`approve_workflow(id, …)` / `reject_workflow(id, …)`	Resolve an approval gate.
`get_approval_audit(approval_id=…)`	Look up an approval decision.
`get_policy_audit(…)`	Look up lifecycle policy-action firings.

`BoundFlowWorker`¶

Register workflow handlers and connect to the backend:

worker = BoundFlowWorker(llm=AnthropicLlmClient(...))

@worker.workflow("triage", version=1)
async def triage(ctx):
    await ctx.run_agent(AgentDefinition(...))
    return Complete()

await worker.run()

Inside a handler, ctx provides add_context(...), run_agent(...), mark_failed(), and follow-on operation registration (e.g. approval branches). Attach a trace_sink= to export run telemetry — see Observability.