Loops

Beta. Loops is available to eligible organizations and is still evolving. The current node catalog and limitations are documented below.

Loops is Fondaro's visual workflow builder. Each Loop is a graph of connected nodes that can gather CRM data, draft work with AI, wait for a person, and run supported CRM actions.

Runs are durable. Closing your browser, restarting the API, or deploying the workflow runtime does not discard a run that is waiting at a human checkpoint. Fondaro keeps the run and node history as its audit trail.

Loops requires an eligible active plan or funded trial. The entitlement is checked when you open the builder and when a manual, scheduled, or event trigger starts a run. The dashboard route and authoring controls are currently available to organization admins.

Create a Loop

Open Loops from the dashboard sidebar and choose one of three starting points:

• Build with Fondaro AI opens the dashboard conversation with a Loop-authoring request. Describe the outcome and let Fondaro AI ask about the trigger, steps, and approval point.
• Start from a template copies one of the working example graphs into your organization.
• Start from a blank loop opens an empty canvas.

Fondaro AI can create a new Loop or update an existing one using the same supported node catalog as the editor. A new AI-authored Loop is paused. Open its inline result card, inspect the graph, and activate it only when the wiring and configuration are right.

Edit the canvas

Use Add node to browse triggers, data, AI, human, action, and display nodes. Drag from an output port to a compatible input port. The editor rejects connections whose data types do not match.

Select a node to configure it in the inspector. Canvas and configuration changes save automatically after a short pause, and the header briefly shows Saved. Editing is locked while the latest run is active and while you inspect a historical run.

See the Node catalog for the 26 current node kinds and their important safety differences.

Activate or run it

A Loop can start manually, on a five-field UTC schedule, or when one of the supported CRM events happens. Scheduled and event Loops only fire automatically while they are Active. Choose Pause to stop new automatic runs. Manual testing remains available from the editor.

Before activating an automatic Loop, use Test this loop to choose a trigger and inspect a real run. See Triggers for schedules, event types, and testing limits.

Human control

There are two human checkpoint nodes:

• Human review pauses the run and asks you to choose which input rows continue.
• Human approval turns upstream AI drafts into individual proposals. You can edit supported fields, approve, or dismiss each proposal. The run resumes after every row has a decision.

AI nodes only create proposals or a structured refusal. They do not write directly to your CRM. Connect their proposal output to Human approval when you want the proposed work to execute.

Action nodes are different. They are deterministic CRM mutations and run as soon as the graph reaches them. They are not automatically wrapped in a human checkpoint. Only add a direct action when that behaviour is intentional, and use the editor's typed connections to keep its input scoped.

Follow a run

The canvas shows each node as pending, running, awaiting input, complete, skipped, or failed. Select a node to inspect its input, output, timestamps, or error. The bottom run strip lets you reopen recent runs without leaving the editor.

See Runs for approvals, history, failure behaviour, and cancellation.

Current limitations

• Loops is in Beta and the catalog is fixed to the 26 documented node kinds. There is no wait or delay node yet.
• The Research agent currently uses scoped CRM and connected-property searches. It does not perform general web research or return web citations.
• The Brochure generator currently prepares a brochure outline as a note proposal. It does not render a PDF.
• Schedules use UTC and the dispatcher checks a five-minute window, so Loops is not a sub-five-minute scheduler.
• Event bursts are deduplicated and capped. A single source event can start at most 25 runs, and an organization can start at most 100 event runs per rolling minute.
• A failed node cannot be restarted in place. Fix the graph or source data, then start a new run.