Operational modes · you choose the autonomy

Run it your way — human, hybrid, or autonomous.

Scopes runs in three modes so you decide exactly how much an agent does on an arc — from pure observer to acting within policy. One toggle per scope. Change it any time. Provenance is preserved regardless.

human-led

agent autonomy →

traditional

Traditional

You do everything.

agent autonomy

agent can

Read arc history
Surface context

human must

Log every beat
Draft every message
Make all decisions

observer

Observer

Agent watches. You act.

agent autonomy

agent can

Watch beats
Surface suggestions
Flag risks
Draft context docs

human must

Approve every action
Send every message

collaborative

Collaborative

Agent acts. You approve risk.

agent autonomy

agent can

Act within policy
Log beats automatically
Escalate high-risk
Draft + send within bounds

human must

Approve high-risk actions
Set policy bounds

mode is configurable per scope · per org · per risk threshold

The spectrum goes from fully human-controlled to agent-acts-within-policy. Every action at every level is signed, attributed, and permanently on the arc.

Modes configure where the agent stops.

Traditional

The agent is a reader.

No actions. No suggestions unless asked. Agents can read the arc and surface context to humans — nothing more. Useful for teams that want provenance without changing their workflow.

Observer

The agent is a co-pilot.

The agent watches every beat and surfaces drafts, alerts, and recommendations — but never acts. Humans see the suggestions and choose whether to apply them. A low-friction way to try AI assistance.

Collaborative

The agent is an executor.

The agent acts within the bounds you set: message, log, escalate, close. High-risk actions — above your configured risk threshold — are paused and surfaced for human approval before they run.

High-risk still asks a human

Collaborative doesn't mean unsupervised.

You set a risk threshold per scope — or per org. Any proposed action above it is queued for human approval before it runs. The agent stays fast on routine work and defers on the actions that matter most.