Flitch activity
Chart Flitch itself: who reads your dashboards, what is failing, and where the spend goes.
Every other source tells you about your business. This one tells you about your Flitch.
Add it and you get a dataset you can chart like any other, combine with your own data, or ask the assistant about. There is nothing to connect: it is your own workspace, so there are no credentials.
Objects
Tick as many as you want when you add the source. Each becomes its own dataset.
| Object | One row per | Columns |
|---|---|---|
| Dashboard usage | dashboard, day and viewer | dashboard_id, dashboard, day, viewer_id, viewer, views |
| Refresh history | dataset and day | dataset_id, dataset, connection, day, runs, failures, rows_processed, avg_duration_ms, last_error |
| Credit usage | day, person, type, area and model | day, user, type, area, model, credits, cost_usd, transactions |
| API usage | caller, endpoint and day | day, caller, caller_type, endpoint, requests, errors, rate_limited |
| Dataset usage | dataset | dataset_id, dataset, source, rows, last_refreshed_at, refresh_interval_minutes, used_by_dashboards |
Dataset usage is current state rather than history, so the date window does not apply to it. used_by_dashboards counting zero is the interesting case: nothing in Flitch otherwise tells you which datasets nothing reads.
API usage names a key by its name and a person by their email, so caller_type tells you which. A machine key and a signed-in person fail and get throttled for quite different reasons.
In credit usage, type separates spending (usage) from topping up (allocation, purchase), and credits is positive in both cases, so filter by type before summing rather than relying on the sign. area says what the spend was for: building a dashboard, the data assistant, an API call.
Ids sit alongside names because names are not unique. Two dashboards can share a title, and so can two people, so grouping a chart by name alone can quietly sum things that are unrelated.
Add
Open Data
In app.flitch.io, open the Data page.
Add the source
Top right, click Add, then pick Flitch activity under Quick start.
Choose what to include
Tick any of the five, and choose how far back to report: 30 days, 90 days, or a year.
Create
Each one you ticked becomes its own dataset, named after what it holds, and appears in your library straight away with rows already in it. Rename any of them from the library if you want something else.
Refresh
Nothing to schedule. The rows are recomputed every time the dataset is read, so a chart built on it moves as people use your dashboards. There is no cached copy to go stale and no refresh to fail.
What is not included
Aggregates only, at every grain.
People signed in to Flitch are named by email. Everyone reading a public link counts as a single Anonymous row per day, because there is no account behind that visit. The IP address, browser and city recorded against a public view are never included, in the dataset or through the API.
Refresh history is aggregated by day rather than per run. The underlying log runs to millions of rows, so a row per run would be unreadable as well as enormous, and the day grain answers which pipelines are failing without the noise.
Ask instead
The assistant on the Data page can read all five without you adding anything, so you can just ask:
- Which dashboards has nobody opened this month?
- Show refresh failures by dataset over the last 30 days
- Which datasets is nothing using?
- What is using most of our credits?
Through the API
The same five objects are available at GET /api/v1/space/{object}, reading the same data. See the API reference.
Troubleshooting
No rows in dashboard usage. Only published dashboards are counted, and only real viewers. Your own editor sessions are excluded on purpose, so a dashboard you have built but not published shows nothing.
Refresh history is empty. It covers connections that run on a schedule. A dataset you uploaded once has nothing to refresh, so it never appears.
A dashboard shows as "Untitled dashboard". It has no name saved. Rename it in the editor and it will be named here on the next read.