Workflows

Shows how to run deterministic multi-step workflows from inside an agent using a custom run_workflow tool. The LLM decides when to call the tool and maps user intent to workflow_name and input; your orchestration engine runs the steps in order, retries on failure, and returns a structured result. Source: examples/agent_with_workflows/

What it demonstrates

Deterministic workflow execution — steps defined in your engine, not improvised by the LLM
WorkflowTool + WorkflowRunner interface — pluggable engine (in-process, Temporal, Conductor, Argo, etc.)
Agent waits until the workflow finishes — Execute blocks on WorkflowRunner.Run
Long-running config — AgentModeAutonomous, WithTimeout, WithToolExecutionConfig sized for workflow duration

Run

From examples/: In-process runner — no infrastructure needed:

go run ./agent_with_workflows "Run the onboarding workflow for user Alice"

Temporal runner — durable, survives restarts:

task infra:temporal:up && task infra:temporal:wait
ORCHESTRATION_ENGINE=temporal go run ./agent_with_workflows "Run the onboarding workflow for user Alice"

Requires LLM_APIKEY in examples/.env.

Key code

// newRunner picks the engine from RUNNER env var (default: in-process).
// Swap inProcessRunner for NewTemporalRunner, a Conductor runner, etc.
runner, stop, err := newRunner(cfg)
defer stop()

a, err := agent.NewAgent(
    agent.WithLLMClient(llmClient),
    agent.WithTools(NewWorkflowTool(runner)),

    // Workflows can be long-running — autonomous mode + explicit tool ceiling.
    agent.WithAgentMode(agent.AgentModeAutonomous),
    agent.WithTimeout(10 * time.Minute),
    agent.WithToolExecutionConfig(agent.ExecutionConfig{
        Timeout:     10 * time.Minute,
        MaxAttempts: 1, // do not blindly re-trigger a started workflow
    }),
)

agentRun, err := a.Run(context.Background(), prompt, nil)
if err != nil {
    log.Fatal(err)
}
result, err := agentRun.Get(context.Background())

Execute blocks on WorkflowRunner.Run — the agent run stays open until the workflow finishes or a timeout fires.

Expected output

user: Run the onboarding workflow for user Alice

assistant: Onboarding for Alice is complete. A welcome email was sent and a workspace was provisioned.

Before You Begin

Basic Agents

Configuration

Tools & MCP

Streaming & AG-UI

Memory & RAG

Observability & Hooks

A2A

Multi-agent

Durable & Distributed

What it demonstrates

Run

Key code

Expected output

Learn more

Deterministic Execution

Execution config

​What it demonstrates

​Run

​Key code

​Expected output

​Learn more

Deterministic Execution

Execution config

What it demonstrates

Run

Key code

Expected output

Learn more