Skip to content
For the complete documentation index, see llms.txt. This page is also available as Markdown.

Human-in-the-Loop

The points where a person stays in the decision.

An agent that acts on its own is the point. An agent that acts on its own where it shouldn't is the risk. In a bank, the difference is usually not whether the agent is capable — it's whether a person is standing in the right place when it commits to something.

Qore's controls fall into two families, and it's worth being precise about which is which:

  • Automated controls — Guardrails & Validation such as Sensitive Data Masking and model policy, and Evals scoring live traffic. A machine checks the agent.

  • Human controls — the subject of this page. A person checks the agent.

Where a person can stand

Before the agent commits: Autonomy

Autonomy is a dropdown — Low, Medium, High, defaulting to Medium. It sets how independently the agent acts without checking back.

  • High — the agent commits to multi-step plans on its own.

  • Low — the agent confirms more steps before proceeding.

This is a dial rather than a switch.

Before a skill runs: Tool Execution Approval

Tool execution pauses the agent before it runs selected skills and asks the user whether the call should proceed. Allow runs the skill and continues the run. Deny blocks the call, and the agent continues without it. Choose the skills one by one, or turn on Enable for all skills to cover every current and future skill.

When an agent needs an answer: questions and Gates

In an Operator conversation, an agent that needs a decision stops and asks. The question appears as a card above the composer. Pick a numbered answer, write your own, attach a file when the agent asks for one, or select Skip.

A conversation that ends on a question waits until you answer. It is listed on Gates, in the top bar of Console and of Control Center. Only your requests appear there, and only you can answer them.

Before a governance change runs: staged actions

The Governance Operator in Control Center doesn't act on its own. It stages each proposed change, and nothing runs until you approve it. The Console Operator, by contrast, dispatches work as soon as you ask.

When an evaluator is unsure: Review Queue

An evaluator scores each live execution with a rating and a confidence. When the confidence is below the evaluator's threshold and routing to human review is turned on, the result becomes a task in Review Queue, where a reviewer records a verdict. See Evals.

Before the agent learns: learning-case approval

The agent can improve from its own production history — but nothing enters its memory without a person approving it.

Before code exists: plan review

In App Builder, Plan Mode presents a proposed architecture — features, logic, extra ideas — and waits. You click Start building to commit, or Revise plan to iterate. Nothing is generated until you agree. See Plan.

Seeing enough to intervene

A person can only stay in the loop if they can see what the agent did:

  • The execution trace — in the builder, the full record of how the agent read your instructions, which skills it chose, and how long each took. For builders, not end users.

  • The Operator transcript — each step the Operator took, each question it asked with the answer beneath it, and a card for each agent it handed work to. Open full run shows every step that agent took. See Console Operator.

  • Monitor and Evals — what the agent actually did in production, scored.

The human controls at a glance

Control
Where it is set
What the person does

Autonomy

Agent Config → Agent Behaviour

Sets how many steps the agent confirms

Tool Execution Approval

Agent Config → Guardrails

Allows or denies each selected skill call

Questions and Gates

Operator conversations; Gates in the top bar

Answers, attaches a file, or skips

Staged actions

Governance Operator, in Control Center

Approves a proposed change before it runs

Review Queue

Control Center → Evals

Records a verdict on a low-confidence score

Learning-case approval

Agent Config → Reinforcement Learning → Learning Cases

Approves a correction before it enters memory

Plan review

App Builder, Plan Mode

Starts building or revises the plan

Where this is configured

Last updated

Was this helpful?