HITL

Create review loops on the HITL.sh platform, send content to human reviewers, and resume once a person responds. Agents use it as an external review surface when the reviewer already works in HITL.sh.

1 trigger 14 actions API key available

HITL website ↗ Platform Documentation ↗ Capability data verified 2026-08-01

A moderation agent flags an item its classifier scored in the uncertain band

Agent assembles the item, the classifier score and the rule that was nearly matched

Agent calls Get Loop and confirms the on-call moderation loop still has active members

Agent calls Create Review Request as a time sensitive single select, and every active member gets a push notification

Agent records the request ID against the item so the queue can be reconciled later

On Request Reviewed fires with the reviewer's verdict, who answered and how long they took

A moderator answers from the HITL mobile app, and the agent applies their verdict rather than the classifier's guess

What This Integration Enables

HITL.sh is organized around the loop: a named container that binds a group of reviewers to the requests they receive. Reviewers join by invite code or QR from the mobile app, and every active member of a loop is pushed a request the moment it is raised. That mobile-first shape is what the connector is really for. If the decisions you need are the kind somebody has to answer while walking between meetings, a queue that lives on a phone gets answered and a queue that lives in a browser tab does not. The connector manages loops and their membership, raises and cancels requests, reads results back, and records feedback on how well a reviewer performed.

The response types are the second reason to reach for it. A request can ask for free text, editable text, a single select, a multi select, a rating, a number or a boolean, so one connector covers an approve-or-refuse gate, a five-point quality score on model output, and a draft the reviewer rewrites in place. Requests can be marked time sensitive with a timeout between one minute and twenty-four hours, after which HITL.sh applies the Default Response automatically.

FlowRunner's human-in-the-loop is a native platform capability: an agent can call a pause as a tool from inside its own reasoning, reach a person on email, Slack, WhatsApp or phone, and pick the run back up with its whole context intact. That is the default and it goes deeper. This connector exists because your reviewers may already carry the HITL mobile app, and FlowRunner should not ask them to install a second thing to answer the same kind of question. The reviewers stay where they are. The connectors, the orchestration and the decision about when to stop stay in FlowRunner.

Without FlowRunner

Reviews wait for a desk An urgent verdict sits until the reviewer is back at a laptop

Approve or refuse only A draft that is nearly right gets rejected instead of fixed

Unowned queues Nobody notices a review group has drifted down to zero active reviewers

With FlowRunner

Answered from a phone The request arrives as a push notification and the verdict comes back in minutes

Seven response shapes The same connector covers a yes-no gate, a quality rating and an edited draft

Coverage checked first Reviewer coverage is confirmed before a time sensitive request is dispatched

Use Case Scenarios

On-call exception review with a safe default

An overnight processing agent hits an exception it cannot classify. It raises a time sensitive request to the on-call loop with a single select of the three actions it could take, and a Default Response of "hold for morning". Every active member of the loop is pushed the request. If someone answers, the agent takes the chosen action immediately. If nobody answers before the timeout, the default applies, the item is parked, and the agent moves on to the rest of the batch rather than stalling the whole run. The overnight work finishes either way, and nothing was decided by guessing.

Moderation queue with reviewer scoring

Flagged content is routed to a moderation loop as a single select verdict with an optional note. Get Review Request returns the answer along with who responded, how long they took and which loop members were notified. A weekly flow calls List Loop Review Requests to pull the whole queue for the period, and Add Request Feedback to record a rating on the calls that turned out to be wrong. Over time the loop's roster gets managed on evidence rather than impressions, and the reviewers who are fast and accurate are the ones the urgent work goes to.

Keeping the reviewer roster in step with the org chart

An HR-driven flow watches for leavers and role changes in BambooHR. When someone leaves the support organization, the flow calls List Loop Members to find every loop they belong to and Remove Loop Member on each one, so they stop receiving requests on their phone. Requests the person has already claimed are untouched by that removal, so the same flow calls Cancel Review Request for anything still in flight and re-raises it against the loop. The roster stops drifting away from reality, which is the failure that makes a review queue useless without anyone noticing.

Human-in-Loop Highlight

A review product's worst failure is not a bad decision. It is a decision that was never made and gets recorded as one anyway. HITL.sh makes that possible in a specific, documented way: a time sensitive request applies its Default Response automatically when the timeout expires, and On Request Reviewed then fires with a final state that looks exactly like a human answer to any flow that is not paying attention. Combine that with a loop whose members have quietly gone to pending or been removed, and an agent can run an entire night of "approved" refunds that no person ever saw.

So FlowRunner puts the gate in front of dispatch. Before raising a time sensitive request whose Default Response is anything other than a genuine no-op, the agent calls Get Loop and List Loop Members and checks for active reviewers. If the loop has none, it does not send a request into an empty room. It stops and messages the operations owner: "The payments-oncall loop has zero active members. Three refund approvals are queued and their default is Approve on timeout. Reassign to the finance loop, change the default to Hold, or send anyway?" The person answers, the flow resumes with their instruction, and the difference between an approval and a timeout stays visible.

Agent processes routinely

Detects exception requiring judgment

Clear match Continues automatically

Ambiguous Routes to human via preferred channel

Human decides

Agent resumes with decision

Agent Capabilities

14 actions

Account

Test API Key Verifies that the configured HITL.sh API key is valid and returns the account details, the key identifier and the current rate-limit status. Used as the first connectivity check when standing up the integration, and as a diagnostic when every operation starts failing at once, since they all share one credential.

Loops

Create Loop Creates a new review loop, the container that binds a group of reviewers to the requests they receive. The caller becomes its first member, and the response includes an invite code, a join URL and a QR code that reviewers use to join from the mobile app. Used when a flow needs a review group that did not exist before, such as a queue for a new product line.
List Loops Retrieves every loop created by the authenticated account with member counts, pending invitation counts and creation timestamps. Used to discover the loop identifier that Create Review Request needs, and to report on how thinly reviewers are spread across queues.
Get Loop Retrieves a single loop by identifier, including the full member list, member statistics and status. Used as the pre-flight check that a loop still has live reviewers before an automation dispatches work to it.
Update Loop Updates the name, description or icon of an existing loop. Only the loop creator can perform the update, and any field left empty is preserved. Renaming a loop leaves its invite code, members and in-flight requests alone.
Delete Loop Permanently deletes a loop together with all of its membership and request history. Only the creator can delete a loop and there is no undo, so on a shared loop this deserves an approval of its own before the call is made.
List Loop Members Retrieves every reviewer attached to a loop with their identifier, email, membership status and join date. Used to confirm coverage before sending time-sensitive work, and to build a roster report of who is actually able to act on approvals.
Remove Loop Member Removes a reviewer from a loop so they stop receiving new requests on their device. Only the creator can remove members, and requests the reviewer has already claimed are unaffected, so pair this with Cancel Review Request when something in flight also has to be withdrawn.

Review Requests

Create Review Request Sends content to a loop for human review and immediately pushes a notification to every active member. Used at the moment an agent reaches a judgment call: a verdict on flagged content, a rating on model output, or a draft for a person to polish. The action returns as soon as the request is queued rather than waiting for an answer, so pair it with the On Request Reviewed trigger or supply a callback URL. The expected answer is shaped by Response Type together with Response Config, and a time sensitive request needs a timeout after which the Default Response is applied.
List Review Requests Retrieves requests created by this API key, filterable by status, priority and loop, each with its current status, the reviewer's answer and the response time. Used to page through a backlog or to build a dashboard of outstanding approvals.
List Loop Review Requests Retrieves every request belonging to one loop regardless of which API key raised it. Used to audit a single approval queue end to end, where List Review Requests would only show the flow's own share of it.
Get Review Request Retrieves one request by identifier, including the answer, who responded, how long they took and the broadcast result showing which members were notified. Used when a flow raised a request earlier and now needs to check whether a person has decided.
Cancel Review Request Withdraws a request that has not been completed, removing it from reviewers' queues and stopping its reminders. Used when the work behind a request is cancelled or superseded. Only pending or claimed requests can be cancelled, so a request a reviewer has already answered returns an error rather than silently disappearing.
Add Request Feedback Records feedback on a request a human has already completed, so reviewer performance can be scored and quality trends surfaced. Used in the weekly loop that turns review outcomes into roster decisions. Feedback is free form, with a rating and a comment as the conventional fields, and can only attach to completed requests.

Triggers

1 triggers

Event Triggers

On Request Reviewed Fires once a review request reaches a final state, whether a reviewer completed it, it timed out and fell back to the default response, or it was cancelled, so a paused run can resume with the human decision in hand. Each event carries the reviewer's answer, the responding user, the response time and the original request text, and can be narrowed to a single loop or to one outcome. Because a timeout arrives here as a final state too, branch on the outcome rather than assuming an event means a person answered.

Frequently Asked Questions

What can FlowRunner do with HITL?

FlowRunner agents can run Test API Key, Create Loop, and List Loops in HITL, plus 11 more actions.

Does connecting HITL to FlowRunner require OAuth?

No. HITL connects to FlowRunner with an API key, no OAuth flow required.

Can HITL trigger a FlowRunner workflow automatically?

Yes. HITL supports 1 trigger that can start a FlowRunner workflow automatically.

Start building with HITL

$100 in credits. No card required. Connect in minutes.

Start Building Free Talk to us