aiKonnector
ProductivityPush an item into an aiKonnector approval or review queue and read the decision back. Agents use it as an external checkpoint when a review already lives in aiKonnector rather than in the flow itself.
What This Integration Enables
aiKonnector is one action wide, and that is the point of it. It is a drop box for a decision rather than a decision system: an agent hands it a JSON object, aiKonnector renders that object on a review screen automatically, and a person opens the URL to approve, reject or edit. There is no form to build and no schema to register first. For teams whose reviewers already keep an aiKonnector queue open, that is the shortest possible distance between an agent hitting an automation exception and a qualified human seeing it.
FlowRunner's human-in-the-loop is native. An agent pauses mid-run, reaches a person on email, Slack, WhatsApp or phone, and resumes with the full context of the run intact. This connector is for the case where the review already lives in aiKonnector and somebody already owns that queue, so there is nothing to gain by moving them. FlowRunner still decides when a run is uncertain enough to stop, still holds the paused run, and still carries the context. aiKonnector is where the question gets displayed. Note the shape of the handoff: Create Approval Request returns the moment the item is queued and does not block waiting for a verdict, and aiKonnector publishes no queue-polling API, so the decision has to travel back to the flow through a FlowRunner webhook that the reviewer's action calls. Design the return path first, then the request.
Without FlowRunner
With FlowRunner
Use Case Scenarios
Expense reports above the policy threshold
An agent runs the nightly expense batch. Reports that pass policy post straight through. Reports with a line above the threshold, or a receipt that does not match the claimed amount, get assembled into a payload with the requester, the offending lines and the policy rule that flagged them, and pushed into the finance review queue with Create Approval Request. The controller opens one queue in the morning instead of one mail thread per report, and each decision releases the corresponding posting in QuickBooks Online or NetSuite. The controller is deciding on exceptions, not sorting a batch.
Compliance sign-off on generated documents
A document generation agent drafts a customer-facing agreement from a template and a set of deal terms. Before the file is sent anywhere, it goes into the legal review queue as a payload carrying the drafted clauses, the terms they were derived from, and the clause the agent changed from the standard. Counsel opens the review screen, edits the wording if needed, and approves. Only then does the flow route the file for signature and file it in Google Drive. The generated document never reaches a customer on the agent's judgment alone.
Spot-checking a newly deployed agent
A team turns on a new classification agent and does not yet trust it at full autonomy. Rather than run it in shadow mode and read logs later, they route a sample of its output into an aiKonnector queue: the input, the classification and the confidence, one item at a time. A reviewer confirms or corrects each one. As the correction rate falls, the team narrows the sampling rule in the flow until only genuinely ambiguous cases are queued. The review volume shrinks because the flow changed, not because the reviews got skipped.
Human-in-Loop Highlight
aiKonnector renders whatever JSON you send it, which is what makes it fast to adopt, and it displays only two levels of object nesting. Those two facts together create the moment worth gating. An agent that assembles a payload from three systems can easily nest the number the decision actually turns on at level three, and the reviewer then sees a screen that looks complete and is not. There is no error, no warning, and no trigger coming back to tell the flow that the approval was granted against a partial view. So the gate sits before dispatch. The first time a flow sends a new payload shape, FlowRunner pauses and messages the process owner in Slack with the flattened two-level view exactly as aiKonnector will render it: "Reviewers will see these 9 fields for expense report 4471. The policy breach amount is nested under lines.policy and will not display on the review screen. Send as-is, or hold while the payload is flattened?" The owner answers once, the flow resumes, and every run of that shape afterwards goes straight through. The agent does the assembling. A person confirms that the reviewer can actually see what they are being asked to decide.
Agent Capabilities
1 actionsApprovals
1- Create Approval Request Adds an item to an aiKonnector approval queue and returns the review URL a human opens to approve, reject or edit it. Used at the point where an agent has finished its work and reached something it should not decide alone: an expense above threshold, a generated contract clause, a customer record about to be created downstream. The action returns as soon as the item is queued rather than blocking on the verdict, so pair it with a notification carrying the returned URL, or let aiKonnector notify the queue's members, and collect the decision through a FlowRunner webhook. The default queue is set on the connection and can be overridden per action when one flow feeds several queues.
Frequently Asked Questions
What can FlowRunner do with aiKonnector?
FlowRunner agents can run Create Approval Request in aiKonnector.
Does connecting aiKonnector to FlowRunner require OAuth?
No. aiKonnector connects to FlowRunner with an API key, no OAuth flow required.
Can aiKonnector trigger a FlowRunner workflow automatically?
aiKonnector doesn't currently expose triggers in FlowRunner. It connects as an action step inside workflows started by another trigger.
Start building with aiKonnector
$100 in credits. No card required. Connect in minutes.