Cyfe
Analytics & DataConnect AI agents to Cyfe, an all-in-one KPI dashboard platform. Agents push metrics from anywhere in a flow onto a Cyfe widget so a dashboard reflects live operational state.
What This Integration Enables
Cyfe is a dashboard, and this connector treats it as exactly that: an output surface, not a data source. The Push API accepts data into a Custom Push API widget and nothing else, so the useful pattern is not "read metrics from Cyfe," it is "the flow that does the work also reports the number." That turns out to be the more valuable direction. The counts operations teams actually care about, such as how many invoices needed a human this week, how many orders cleared without touching a person, how many exceptions are still open, are numbers that only exist inside the flow that produced them. Nothing else in the stack can compute them, because no single system saw the whole run. FlowRunner agents push a single named metric with an optional dimension label, chart type, colour, and duplicate handling, or push one or more data rows carrying multiple metrics each, with per-metric maps for chart type, colour, duplicate handling, cumulative totals, and average header values.
Three constraints define what you can and cannot build here, and they are unusual enough to plan around. First, authentication is per widget rather than per account: adding a Custom Push API widget generates a unique push URL whose embedded token is both the endpoint and the credential, so each connection targets exactly one widget and you create a separate connection for every widget you push to. Second, the API is push-only. There is no public read or query API for listing dashboards or widgets, and no delete, so a flow can write to a widget and can never read back what the widget currently holds or remove what it wrote. Third, there are no triggers, and there is no polling alternative either, because there is nothing to poll. Cyfe never starts a flow; the flow starts elsewhere and ends here. The practical consequence of the second and third points together is that On Duplicate matters more than it looks: Add is the default and sums repeated dimensions, Replace overwrites them, and since you cannot inspect the current value, choosing wrong is not visible until someone questions the chart. Cyfe enforces a monthly limit of 100,000 push requests per account. The connector is built and verified against Cyfe's official Push API.
Without FlowRunner
With FlowRunner
Use Case Scenarios
The Exception Rate Nobody Could Measure
An accounts payable flow processes invoices and routes the ones it cannot resolve to a human. At the end of each run, the agent pushes two metrics to an operations widget with Push Metric: the count that cleared automatically and the count that needed a person, both labelled with the day's YYYYMMDD dimension and given different chart types and colours so they read as two series. Over a quarter the widget shows the ratio moving, which is the number that tells an operations lead whether the automation is actually getting better or just getting busier. No other system in the stack could have produced it, because no other system saw both halves of the run.
A Live Fulfilment Board
A logistics flow moves orders through validation, allocation, and dispatch, drawing on several systems along the way. Rather than exporting from each of them overnight, the agent calls Push Data Points at the end of each cycle with one row carrying orders received, orders dispatched, orders held for review, and average hold duration, and passes per-metric maps so hold duration renders as a line while the counts render as columns, and so the cumulative header shows the running day total. A warehouse lead watching the board sees the shift as it happens rather than as yesterday's summary.
Reconstructing History After a New Widget
A team adds a new widget for a metric they have been computing for months without displaying. The agent reads the historical values from the store they were logged to, such as a table in Google Sheets, and calls Push Data Points with one row per historical day, each carrying its YYYYMMDD label, with the On Duplicate map set to replace for every metric in the payload so a re-run corrects rather than compounds. Because a backfill of this shape cannot be checked afterwards, the shape of the payload is confirmed before it runs, not after.
Human-in-Loop Highlight
The consequential moment in a two-action connector is not a delete, because there is no delete. It is On Duplicate. Push Metric and Push Data Points both accept duplicate handling, and the default is Add, which sums repeated dimensions rather than replacing them. That default is correct for the common case, an incremental counter pushed once per cycle, and quietly wrong for the case that matters most: a historical backfill, or a re-run after a failure, where the same YYYYMMDD label is sent twice. Push a month of history with the default and every day that already had a value now shows double it. What makes this worth a human decision rather than a retry policy is Cyfe's own shape: there is no public read API to check what the widget currently holds, and no delete to remove what was written. The mistake is invisible from the flow and permanent from the flow, and the first person to notice is whoever is reading the chart in a meeting. So the agent gates the backfill, not the routine push. Before any run that sends more than one dimension label, or that sends a label it has sent before, it posts to the dashboard owner: "Backfill to widget [name]: [n] rows spanning [first label] to [last label], metrics [list]. On Duplicate is set to [value] for each. This widget cannot be read back or cleared through the API, so a wrong setting is not recoverable from the flow. Run, change duplicate handling, or cancel?" The owner confirms. The daily push that follows a normal run needs no approval, because it is one label, once, exactly as designed. That is human-in-the-loop placed on the irreversible operation rather than on the routine one.
Agent Capabilities
2 actionsPush Data
2- Push Metric Pushes a single named metric value to the connection's widget, with an optional dimension label such as a YYYYMMDD date, an optional dimension column name that defaults to Date, chart type, hex colour, and On Duplicate handling of Add or Replace. Best for a simple one-value KPI update at the end of a flow run.
- Push Data Points Pushes one or more data rows, each of which can carry several metrics, with optional per-metric maps for chart type, colour, duplicate handling, cumulative header totals, and average header values. Best for multi-metric cycles and for historical backfills, where the On Duplicate map is the setting that decides whether a re-run corrects or compounds.
Frequently Asked Questions
What can FlowRunner do with Cyfe?
FlowRunner agents can run Push Metric and Push Data Points in Cyfe.
Does connecting Cyfe to FlowRunner require OAuth?
No. Cyfe connects to FlowRunner with an API key, no OAuth flow required.
Can Cyfe trigger a FlowRunner workflow automatically?
Cyfe doesn't currently expose triggers in FlowRunner. It connects as an action step inside workflows started by another trigger.
Start building with Cyfe
$100 in credits. No card required. Connect in minutes.