Skip to main content
sureops has two control modes that determine how much autonomy the AI exercises during an active incident. You can override the mode per-incident (and per-stage) at any time, and you can configure the default per environment and per organization. A Maximum auto-executed risk ceiling (max_auto_risk) and an auto-approve toggle further tune what Self-driving mode is allowed to do without pausing — see How the default mode is configured below.

​
Supervised mode

In Supervised mode, sureops drives the incident pipeline but every batch of recommendations pauses for your approval before anything executes. Detection and triage run automatically with no gate; once diagnosis produces recommendations, the AI presents its findings — action, rationale, risk, and evidence — and waits for your explicit approval before resolution runs. Nothing executes until a human explicitly approves. This is the right starting point when you are:
  • Evaluating sureops against your environment for the first time
  • Onboarding a new environment where the AI hasn’t yet built context
  • Running high-stakes production incidents where you want maximum oversight
  • Building team confidence in how the AI reasons about your specific stack
When you approve a recommendation, the AI proceeds. When you reject or modify it, the AI incorporates your feedback and continues with the adjusted approach. Everything you provide — corrections, additional context, chat messages during a stage — carries forward and improves the AI’s reasoning for the rest of the incident.
Supervised mode does not, by itself, pause the incident between every stage — that’s a separate axis. Detection, triage, and (once you’ve approved recommendations) resolution and verification proceed on their own without waiting for you to click through each stage. What Supervised mode does gate automatically is the approval step above, before any recommendation executes. If you want the incident to stop after every stage so you can inspect it before it advances — with Approve & Continue, Re-run Stage, Mark Complete, Rollback, and Skip controls — see Taking and releasing control below: that’s what Take Control gives you, in either mode.

​
Self-driving mode

In Self-driving mode, sureops drives the full incident lifecycle without waiting for you to advance each stage manually. It monitors your systems, detects incidents, runs through detection → triage → diagnosis → resolution → verification → closure, and notifies you when it’s done — or when it hits something it can’t resolve on its own. Self-driving mode does not mean approvals are off by default. Auto-executing a recommendation is opt-in, and there are two separate, mutually exclusive settings that turn it on:
  • Auto-approve agent actions — the most permissive option. Once enabled, sureops approves every pending recommendation automatically, at any risk level, with no risk check applied at all.
  • Auto-execute low/medium-risk fixes — a narrower option, available only when auto-approve is off. sureops auto-executes a recommendation only when it’s at or below your configured Maximum auto-executed risk ceiling (max_auto_risk, capped at “medium” — “high” is never an option for this ceiling). Anything above the ceiling falls back to waiting for your approval.
Until you turn on one of these, the post-diagnosis approval gate always pauses and waits for you — even in Self-driving mode. You remain in the loop at all times:
  • Real-time notifications at key milestones
  • Full timeline visibility — every action the AI took, every tool it queried, and why
  • One-click “Take Control” at any point to pause the AI and step in
  • Two hard carve-outs always require explicit human approval, regardless of mode or auto-approve settings, and neither can be turned off: P1/P2 severity incidents, and any recommendation that opens a fix PR against your repository. Outside of those two, an action’s risk level does not by itself force a pause — auto-approve turns off risk-based gating entirely
Self-driving mode is appropriate once you’ve validated that the AI understands your environment and produces reliable diagnoses and fixes. Most teams reach this point after a few supervised incidents on non-critical environments, then expand the autonomy envelope as trust grows.
P1 and P2 severity incidents are never auto-executed without human eyes — even in Self-driving mode with auto-approve enabled. This is a hard severity gate, not a configurable setting: high-severity incidents always pause for human review before resolution actions run.
The other hard carve-out: any recommendation that opens a fix PR against your repository always pauses for human review before merge, regardless of mode or auto-approve settings. A fix PR mutates your source repository once merged, so you always see the draft diff before approving it.

​
Taking and releasing control

Separately from the autonomy mode described above, you can take direct control of an active incident at any time — this is a different axis entirely:
  • Take Control — immediately signals the AI to finish its current operation and hand the incident to you. The AI doesn’t stop mid-sentence; it completes its current task, saves all findings, and pauses.
  • Release to AI — hands the incident back to the AI. It picks up from the current stage with all context intact, including anything you added while in control.
Take Control is also the gate for the stage-by-stage control panel. Approve & Continue, Re-run Stage, Mark Complete, Rollback, and Skip are not options you can reach directly while the AI is driving — they only appear once you’ve taken control of the incident. This holds regardless of which control mode (Supervised or Self-driving) the org is in.
Taking control does not change the incident’s underlying autonomy mode (Supervised vs. Self-driving) — it only changes who’s driving right now. When you release control, the incident resumes under whatever mode it was already resolved to; the takeover itself never flips that setting.
The auto-release timer (30 minutes) returns control to the AI if you take control and don’t explicitly release it. You can extend the timer by 15 minutes at a time, as many times as you need, while you’re in control; there’s no way to disable auto-release entirely.

​
How the default mode is configured

The default mode is a policy, not a fixed setting. It resolves in this precedence order:
  1. Per-stage override on the current incident (most specific)
  2. Per-incident mode override
  3. Environment default (your staging vs. production environments can have different defaults)
  4. Organization default
  5. System default (Supervised)
You configure the organization default in Settings → Organization, and each environment’s default in Settings → Environments → [environment]. (Settings → AI Agents is a separate page — it configures which LLM model your agents run on, not autonomy mode.) Individual overrides are set from the incident hub.
The trust progression from Supervised to Self-driving is always your decision to make — it doesn’t happen automatically. sureops surfaces confidence scores and audit trails to help you decide when to grant more autonomy, but the configuration change is always manual and always reversible.
Next: Quickstart — evaluate with a demo environment