.png)
Docs / Autonomy & Controls
Autonomy & Controls
An agent that can act on its own is useful right up until it does something you'd never have approved. FluxForce solves that by letting you set the leash, per agent, and by keeping a kill switch you can reach at any moment.
Configurable autonomy
You decide how far each agent can go without a human. The setting follows the risk of the action, not a one-size rule for the whole system.
| Risk of the action | Default behavior | You can change it? |
|---|---|---|
| Low | Agent acts on its own. Routine clears and obvious holds don't need a person. | Yes |
| Medium | Human in the loop by default. The agent recommends, a person confirms. | Yes |
| High | Always a human. This one isn't optional. | No, by design |
The threshold is yours to set, per agent. A cautious team keeps more in the medium band at first and widens autonomy as trust builds. The only fixed line is that high-risk actions always wait for a person.
Human in the loop
When an action needs a person, the agent doesn't just stop and wait silently. It hands over a finished recommendation: the decision it would make, the reasoning behind it, and the context it gathered. The human approves, changes, or rejects it, and that choice is recorded. The agent isn't guessing what you want, and you're not starting from a blank screen.
One control pauses autonomous action. You can scope it to a single agent, a whole squad, or the entire system. When it's pulled, agents stop acting and hand everything back to your team. It's always available, and using it never deletes the evidence already recorded.
Shadow mode
Before any agent acts, you can run it in shadow mode. It makes its call and records the reasoning, but takes no action. You compare its decisions against your team's for as long as you want. When the calls line up, switch autonomy on for that agent. This is how most teams build trust, and it's the reverse of being asked to trust a black box on day one.
Who can change the settings
Autonomy settings, thresholds, and the kill switch sit behind role-based access. Not everyone who can see a decision can change how much an agent is allowed to do. Every change to these controls is itself recorded, so the question "who widened this agent's autonomy, and when" always has an answer.
FAQ
Can we run everything with a human in the loop?
Yes. You can keep every agent in recommend-and-confirm mode for as long as you like. Autonomy is something you grant, not something that's forced on.
What happens to in-flight work when we hit the kill switch?
Autonomous action pauses and pending decisions route to your team. Nothing already recorded is lost, and nothing new acts on its own until you resume.
Can different agents have different autonomy?
That's the point. A fraud agent you trust might run more freely than a newer agent you're still watching. The settings are per agent.
Is the kill switch ever unavailable?
No. It's always reachable. No agent is ever beyond your reach.