NoClick Docs
Sign in

An agent that keeps watch over the app

An agent that keeps watch over what you built in the NoClick app - it hears every failed workflow run and every app agent asking for you, looks into each, answers what it can and pauses a workflow that keeps failing - and your code asking an app agent directly.

pip install noclick
python 08_automations/watch_over_the_app.py
EnvironmentNOCLICK_API_KEY
View the code

The workflows and agents you build in the NoClick app run on their own: a schedule fires, a message arrives, an agent works through it. When one breaks, or an agent needs you, you hear about it as an event. This example lists the agents built in the app, asks one of them something from code, and sets up an ops agent of your own that takes a turn on every failed run and every app agent's question: it reads what happened, answers the agent if it can, and pauses a workflow that keeps failing.

pip install noclick
export NOCLICK_API_KEY=nk_live_...
python 08_automations/watch_over_the_app.py
import noclick

The agents you built in the app

Agent.list gives them after the agents made from code, with origin app; each names its workflow, whether it's switched on, and where to change it in the app.

def list_app_agents() -> list:
    agents = noclick.Agent.list(origin="app")
    for agent in agents:
        state = "on" if agent.enabled else "off"
        print(f"{agent.name} in '{agent.workflow_name}' ({state}): {agent.app_url}")
    return agents

Ask one directly

A message runs a turn on the agent's own step in its workflow, as a message from the app does, and the reply comes back whole. conversation continues the thread.

def ask(agent: noclick.Agent) -> None:
    reply = agent.message("In two lines: what are you working on, and is anything stuck?")
    print(f"\n{agent.name}: {reply.text}")

An agent that keeps watch

ops hears two of NoClick's own events. workflow.run.failed comes once per failing workflow (failures within six hours fold into one), so a workflow failing every minute is one conversation, keyed by the workflow. agent.asked_owner is an app agent messaging you. With NoClick's workflows and agents as its tools it can read a run step by step, pause a workflow, and message the agent back.

def keep_watch() -> noclick.Agent:
    return noclick.Agent.from_name(
        "ops",
        create_if_missing=True,
        instructions=(
            "You keep our automations healthy. When a workflow's run fails, read the run, find the step "
            "that broke and why, and pause the workflow if it keeps failing the same way. When one of our "
            "agents asks for us, answer it if what it needs is in the account; otherwise say plainly "
            "what a person has to do."
        ),
        tools=[noclick.Platform.only("workflows", "agents")],
        triggers=[
            noclick.OnPlatformEvent("workflow.run.failed", conversation="workflow-{data.workflow_id}",
                                    message="A workflow's run failed."),
            noclick.OnPlatformEvent("agent.asked_owner", conversation="asked-{data.agent_id}",
                                    message="One of our agents is asking for us."),
        ],
    )


def main() -> None:
    agents = list_app_agents()
    if agents and agents[0].enabled:
        ask(agents[0])
    ops = keep_watch()
    print(f"\n{ops.name} is keeping watch: {ops.trigger_kinds}")


if __name__ == "__main__":
    main()

Clean up

Listing and reading cost nothing; a message to an agent costs its turn. ops takes a turn only when something fails or asks, and noclick.Agent.from_name("ops").delete() removes it.