What is human-in-the-loop here?

Human review is a first-class runtime primitive, not a workaround. The Runtime treats waiting on a human like waiting on a tool return — same primitives, different latency — so oversight is structural.

See the HITL flow → See ReBAC governance →

What is human-in-the-loop here?

Human-in-the-loop (HITL) is the assignment of a step in a Harness to a human for review, approval, or action. What makes it different in Oraclous is where it lives: humans and Agents are symmetric Actors, and the Harness Runtime treats waiting on a human exactly like waiting on a tool to return — the same primitives, just a different latency. Oversight isn’t a manual side-channel you remember to check; it’s a step in the flow the work already runs in.

Citable answer — What is human-in-the-loop in Oraclous? Human-in-the-loop (HITL) is the assignment of a step in a Harness to a human for review, approval, or action. Oraclous’s Runtime treats waiting on a human like waiting on a tool return — same primitives, different latency — so human oversight is a structural part of the flow, not a manual side-channel. The Runtime enforces HITL gates regardless of prose.

What is human-in-the-loop? → · What is an Actor? → · See the HITL flow →

How does human-in-the-loop work?

The HITL flow is one of the platform’s eight named behaviours, and it runs as a sequence of concrete steps:

1. Gate triggered — the Harness reaches a step that requires a human.
2. Assignee resolution — the Runtime works out which human (or human role) the step belongs to.
3. Task creation — a task is created on the workspace task board, the shared surface where Actors — human and Agent alike — pick up work.
4. Notification dispatch — the assignee is notified.
5. Pause and persist — the run pauses and its state is persisted; nothing burns budget while it waits, and the run survives a restart.
6. Human acts — the person approves, blocks, edits, or supplies the input.
7. Resumption — the Runtime resumes the Harness from exactly where it paused.

Because the gate is coded enforcement, not a polite request to the model, prose can’t route around it. If a manifest’s prose tries to skip a HITL gate, the Runtime enforces the gate regardless — the same two-layer model that backs all of ReBAC governance: governance lives in code, flexibility lives in prose, and code wins. A human at a gate can let a step proceed or block it outright, and the policy set the Harness runs under decides where gates are mandatory.

How a Harness works → · How Actors share a task board → · See ReBAC governance →

Why does human-in-the-loop matter?

The usual way teams add oversight to agentic work is a bolt-on: a webhook, a Slack ping, a dashboard someone is supposed to watch. That’s a side-channel, and side-channels get skipped under pressure. Making HITL a runtime primitive means oversight can’t silently fall out of the flow — the run physically pauses at the gate and can’t continue until a human acts.

This is the page for the operations lead who wants to ship automation without giving up the final say, and for the security and compliance leader for whom an unattended high-stakes action is a non-starter. A human approving inside the same flow the Agents run in — with the pause, the persisted state, and the audited resumption — is the difference between “we have a review process” and “review is structurally enforced.” See how this lands for your team on the operations and regulated & security solutions.

Citable answer — Is HITL a workaround or built in? It’s built in. In Oraclous, human-in-the-loop is a first-class runtime primitive: the Harness Runtime treats waiting on a human like waiting on a tool return, pausing and persisting the run at the gate until the person acts, then resuming. Because the gate is coded enforcement, prose cannot route around it — a human can block a step, and the Runtime obeys.

Frequently asked questions

Q: How do humans approve agent work? A: When a Harness reaches a HITL gate, the Runtime resolves the assignee, creates a task on the workspace task board, notifies the person, and pauses the run with its state persisted. The human approves, blocks, edits, or supplies input; the Runtime then resumes the Harness from exactly where it paused.

Q: Is HITL a workaround or built in? A: Built in. Human-in-the-loop is a first-class runtime primitive — the Runtime treats waiting on a human like waiting on a tool return, same primitives, different latency. It’s not a webhook or a dashboard bolted on beside the work; it’s a step in the same flow the Agents run in, and the run pauses there until a human acts.

Q: Can a human block an agent step? A: Yes. At a HITL gate the human can let the step proceed or block it outright. The gate is coded enforcement, so even if a manifest’s prose tries to skip it, the Runtime enforces the gate regardless — code wins over prose. The work cannot continue past a block without human action.

Q: What happens to the run while it waits for a human? A: It pauses and its state is persisted. Nothing consumes budget while it waits, and because the state is durable the run survives a restart. When the human acts, the Runtime resumes the Harness from exactly where it paused — no lost context, no re-run from the top.

Q: Who decides where a human gate is required? A: The policy set the Harness runs under, plus the Harness’s own orchestration. Policy sets are versioned governance envelopes that, among other constraints, anchor where mandatory gates apply; the Runtime enforces them in code. So whether a step requires a human is governance, not a per-run judgement the model makes on its own.