OData v4 feed available for Power BI, Excel and any OData v4 client.
ingestia.io exposes the published marts of your datalake (*_gold and
*_silver datasets) as an OData v4 feed — the format Power BI and Excel consume
natively, with no custom connector.
| URL | What |
|---|---|
https://ingestia.io/api/v1/odata | Service document (lists the tables) |
https://ingestia.io/api/v1/odata/$metadata | EDMX schema (column types) |
https://ingestia.io/api/v1/odata/<tabela> | Rows of a table (paginated) |
This page is the official Power BI/Excel kit — with ready-made recipes for connection, query folding, and incremental refresh.
Availability#
- Status: available. Availability follows your plan's SLA (99.5% monthly from Growth up).
- Clients: Power BI (Desktop and Service), Excel (Windows and Mac 365), and any OData v4 client.
- Region: Brazil (São Paulo).
Permissions#
- API key authentication (see Authentication); only the workspace owner creates/revokes keys.
- Requires the data export action (the owner's data key passes).
- The key sees all published marts in the workspace; PII columns
never leave through the API (not even in
$metadata).
Setup (Power BI — recommended path)#
Home → Get Data → Blank Query → Advanced Editor:
let
Chave = "SUA_CHAVE",
Fonte = OData.Feed("https://ingestia.io/api/v1/odata", null, [
Headers = [ #"Authorization" = "Bearer " & Chave ],
Implementation = "2.0"
]),
Tabela = Fonte{[Name = "sua_tabela"]}[Data]
in
TabelaImplementation = "2.0" is required: it speaks OData v4 and folds (query folding)
filters/projections to the server. Simple path (Excel/PBI):
Get Data → OData Feed with …/api/v1/odata?api_key=SUA_CHAVE and
Anonymous authentication — keeping in mind that the key in the URL becomes a secret.
Query folding — what runs on the server (pushdown)
| Operation in Power Query | Becomes |
|---|---|
| Choose/remove columns | $select |
Filter with = <> > >= < <= | $filter (eq/ne/gt/ge/lt/le) |
| Combine filters (AND/OR/NOT) | and/or/not |
| Contains / Begins with / Ends with | contains/startswith/endswith |
| Keep top N rows | $top |
| Sort | $orderby |
| Count | $count=true |
A filter the feed can't translate returns HTTP 501 with the list of what
is supported — we never silently ignore a filter (a wrong result is worse
than an error). Simplify the filter or apply it locally afterwards.
Incremental refresh (summary)
Create the RangeStart/RangeEnd parameters (Date/Time type), filter the table with
each [data] >= RangeStart and [data] < RangeEnd, and configure the policy. The
date filter folds to the server and only the window is read (fewer bytes = less
cost). Full recipe in the Power BI/Excel kit.
Validation (how to confirm it worked)#
GET /api/v1/odatareturns200withvalue: [{ name, kind, url }].GET /api/v1/odata/$metadatareturns EDMX XML with the column types.- In Power BI, the source step loads and the "Editor" shows the folded steps (right-click → View Native Query, when available).
Example#
# Service document
curl -H "Authorization: Bearer ing_live_..." https://ingestia.io/api/v1/odata
# Página de uma tabela, com projeção, filtro, contagem e ordenação
curl -H "Authorization: Bearer ing_live_..." \
"https://ingestia.io/api/v1/odata/silver_vendas?\$select=data,total&\$filter=data ge 2026-01-01&\$orderby=data&\$count=true"Paginated response:
{
"@odata.context": "https://ingestia.io/api/v1/odata/$metadata#silver_vendas",
"@odata.count": 12345,
"value": [ /*... */ ],
"@odata.nextLink": "https://ingestia.io/api/v1/odata/silver_vendas?...$skiptoken=..."
}Pagination (server-driven)#
- Pages of up to 1,000 rows with
@odata.nextLink— Power BI/Excel follow the link automatically to the end. Prefer: odata.maxpagesize=Nshrinks the page (ceiling 1,000); responds withPreference-Applied.- Always follow
@odata.nextLinkexactly as received (it carries the opaque$skiptoken). A tampered$skiptokenreturns400. - Maximum depth: 1,000,000 rows per query; beyond that,
400telling you to filter (each page scans the entire table in the datalake).
Cost#
- Charges for the bytes actually processed;
x-ingestia-cost-brlshows the amount (a projection today). Every call is recorded in the audit log (api.odata.read, with bytes, page, and key) — separable from native BI. $count=trueruns an additionalCOUNT(*), which is also counted.- Tip: point incremental refresh at a partition column — the datalake prunes partitions and the bytes read really drop.
Security#
- Read-only, scoped to the workspace (published marts only).
- PII columns don't appear in the feed or in
$metadata. - If the customer's own cloud goes down, the feed returns
503— never an empty collection (which would make Excel read "the table was emptied").
Limits#
- OData subset: no
$expand,$apply,$search,$compute, lambdas (any/all), or arithmetic in filters; text onlycontains/startswith/endswith. Outside the subset = explicit501.$format: JSON only. STRUCT/RECORDcolumns are not published; publish flat columns.- Workspace-level key (no per-member filtering).
- Quota: 120 req/min per key (240/min per workspace), separate from the rest of the
platform — consumption via Power BI doesn't degrade native BI. Exceeded →
429withRetry-After. - With no datalake provisioned (demo), the feed serves clearly simulated data and
$filter/$orderby/$countreturn503— it never pretends it filtered.
Troubleshooting#
| Symptom | Likely cause | What to do |
|---|---|---|
501 on a filter | operation outside the subset | simplify the $filter or filter locally |
400 $skiptoken inválido (invalid) | pagination link was altered | follow @odata.nextLink exactly |
429 | quota exceeded | respect Retry-After; reduce parallelism |
503 on the feed | own cloud down / datalake not provisioned | try again; confirm provisioning |
| Missing column | it's PII or STRUCT | PII doesn't leave through the API; publish flat columns |
| High cost | paginating a large table without a filter | use incremental refresh on a partition column |
Next steps: Authentication · Errors and limits · Troubleshooting.