Autonomy & Trust

Control what Solen heals automatically and what requires your approval.

When an agent fails, Solen can resume it from the last checkpoint on its own, or wait for your approval on every heal action. You control which.

Autonomy modes

Supervised (default)

Every proposed heal — checkpoint resume, step-break injection, tool-loop break — is shown in the UI before anything executes. You approve or reject. The agent does not resume without your explicit confirmation.

Example supervised heal approval:

Agent: research-agent · Run: clx9k2m…
Failure: tool_call_loop · search_web x23
Proposed fix:
  + inject_step_break_after_repeated_tool_call
  + resume_from_checkpoint: step_44
Confidence: 97

[ Approve heal ]  [ Reject ]

Use supervised when: you're onboarding agents to Solen, running high-stakes workflows, or want full visibility into every heal decision.

Autonomous

Solen applies heals automatically when confidence is at or above 70. The agent resumes from its checkpoint without waiting. You receive a notification after the fact. Every action is logged in the audit trail.

Example autonomous heal (no approval needed):

03:42:15  Pattern match found (confidence: 0.97)
03:42:18  Policy check passed · workspace mode = autonomous · proceeding
03:42:21  Applying fix · resuming agent from checkpoint step_44...
03:42:39  Incident resolved · 31s end-to-end · zero human intervention

Use autonomous when: you want hands-off agent operation and you've reviewed the safety model on the Self-Healing page.

Changing autonomy mode

In the dashboard: Settings → Workspace → Agent Policy → Autonomy Mode.

Via API:

PATCH /api/agent-runtime/permissions/:agentId
Authorization: Bearer <api-key>
Content-Type: application/json

{ "healMode": "autonomous" }

Valid values: "supervised" | "autonomous". Applies per agent or workspace-wide depending on your policy configuration.

Plan defaults

The confidence threshold

Every heal diagnosis includes a confidence score (0–100) from LearnShift pattern matching or Runtime Kernel AI analysis. Nexus evaluates this against your autonomy policy:

• ≥ 70 + autonomous mode → heal applies immediately, agent resumes from checkpoint
• < 70 + autonomous mode → falls back to supervised, shown for approval
• Any score + supervised mode → always shown for approval

The 70 threshold is fixed and cannot be lowered. Heals below 70 have a materially higher chance of making things worse. This floor is intentional.

Audit trail

Every heal Solen takes — autonomous or approved — is written to an immutable audit log. View it at Settings → Workspace → Audit Log.

Each entry includes:

The audit log is append-only and cannot be edited or deleted.

What Solen will never do autonomously

Regardless of autonomy mode:

• Delete an agent or its checkpoint history
• Resume from a checkpoint that failed verification without opening a new incident
• Apply a heal with confidence below 70 in autonomous mode
• Attempt more than 3 heals on the same incident
• Create a new incident while an existing unresolved one is open for the same run

Read the full safety model →