Workflows

AgentManager coordinates multiple agents.

from maki import MakiLLama
from maki.agents import AgentManager, WorkflowTask

llm = MakiLLama(model="llama3.2")
manager = AgentManager(llm)

manager.add_agent("Researcher", role="researcher")
manager.add_agent("Writer", role="writer")

workflow = [
    WorkflowTask(
        name="research",
        agent="Researcher",
        task="Find the main public-release risks for this repository.",
    ),
    WorkflowTask(
        name="summary",
        agent="Writer",
        task="Summarize the research into a release checklist.",
        dependencies=["research"],
    ),
]

results = manager.run_workflow(workflow)
print(results["summary"]["result"])

Manager patterns

Method	Behaviour
assign_task()	Route a single task to one named agent
coordinate_agents()	Sequential multi-agent pipeline with optional synthesis step
collaborative_task()	All agents work on the same task independently
run_workflow()	Dependency-aware execution with retries and optional parallel batches

run_workflow() orders tasks topologically, retries failures, can run independent tasks in parallel, and can checkpoint progress so an interrupted run resumes where it stopped.

Verified tasks (maker/checker)

A task with verify= is only COMPLETED once a checker accepts it. A REVISE verdict re-runs the task with the checker's feedback, up to max_revisions. If verification ultimately fails, the task fails and its dependents are skipped. See examples/verified_workflow.py in the repository and maki/agents/verification.py for VerifySpec and agent_verifier.

Human approval

Tasks can be gated on human approval before they run; see maki/agents/approvals.py (ApprovalSpec and the approval stores).