Read first:
README.md. This page completes fields 5–7 of the template for all Data flows.
Cost — cost engine (on demand, by bytes read)#
Cost derives from the bytes READ by the query (not from the rows returned), as in an on-demand billing model by scanned volume. Product constants:
| Constant | Value |
|---|---|
PRICE_USD_PER_TIB | US$ 6.25 / TiB (on demand, Brazil region — São Paulo) |
USD_BRL | 5.2 (estimation exchange rate, configurable) |
MIN_BILLED_BYTES_PER_TABLE | 10 MB per referenced table |
MIN_BILLED_BYTES_PER_QUERY | 10 MB per query |
ingestia.io business rule (different from the market standard): EVERY query is charged — there is no free tier and no query costs R$ 0.00. A billed minimum applies:
billed = max(bytesScanned, tableCount × 10MB, 10MB)
usd = billed / 2^40 × 6,25
brl = usd × 5,2The estimate returns {tib, billedBytes, usd, brl}, with standard pt-BR formatting.
Important: this is the estimation engine shown on screen before running. The amount charged is the one on the invoice.
Cost guard — billable-bytes cap#
Every read-only query to the workspace ALWAYS runs with a cap on volume read — a billable-bytes cap that aborts the query before it blows the budget:
MAX_BYTES_BILLEDdefault 1 GiB (an environment setting can override it).- Applied on every run (interactive, parallel verification run, drill), in the Brazil (São Paulo) region and with a workspace label on the bill to attribute the cost.
- Queries are read-only and use named parameters (anti SQL-injection) on the text-to-SQL path.
Limits and caps (concrete)#
| Where | Limit |
|---|---|
| Connector pagination | 10,000 pages · 5,000,000 rows (hard cap, anti-loop) |
| SaaS→Bronze load | default batchSize 5.000 · maxRows 200.000 |
| Profiling | 50 columns · 5 examples × 60 chars · payload ≤ 32 KiB |
| Datalake query | cap on volume read (default 1 GiB) · default timeout 120 s |
| Incremental by period | default window 7 days · cap 400 |
| Incremental cursor by key | literal ≤ 64 chars, no quotes/slash/line break |
Security — per-workspace scope (inviolable invariant)#
- Datalake per workspace: datasets
<prefix>_bronze/_silver/_gold; landinglanding/{prefix}/...in the file area. Theprefixderives from the workspace and is sanitized to[a-zA-Z0-9_-]. A customer never crosses into another's dataset/bucket. BYOC (customer's own cloud) resolves the bucket/service account of the workspace itself. - Secret redaction: every connector error/log has the secret erased — raw, inside a URL (percent-encoded) and in JSON (multiline PEM key). It is the only form of error the adapter lets escape.
- Production does not pretend: a path that would fall into a mock/no-op fails (503) in production instead of simulating success. E.g.: Google Ads is refused in production if the developer token is not approved.
- PII: PII column → examples
•••, min/max/distinct/length suppressed; a hidden column does not even appear in the profile. - Access check: expensive actions (refresh, export, AI, ingestion, API) go through the access check (active workspace). 11 capabilities have a runtime kill switch, including ingestion runs, data console queries, data sampling, dashboard refresh.
Availability and region#
| Item | Value |
|---|---|
| Datalake and processing region | Brazil — São Paulo |
| Contracted availability (SLA) | 99.5% per month from the Growth plan, with service credit — see /legal/sla |
| Starter | best effort |
| Internal target — application availability | 99.9% per month |
| Internal target — pipeline run success | 99.0% per month |
| Internal target — dashboard query (p95) | under 8 seconds |
| Public measurement | /status, with the last 90 days |
| New connectors | start in Limited access; the first run with a real credential is followed by our team |
Fact sheet (9-field template)#
- State — available. Cost engine, guards and per-workspace scope covered by automated tests.
- What it does — estimates cost by bytes read, limits spending per query, enforces caps and isolates each workspace.
- How it works — the cost engine estimates; the billable-bytes cap aborts; sanitized prefix + per-prefix datasets/buckets isolate; secret redaction and the production guard close off leaks and mocks.
- Plan · Permission — core · access check + per-plan entitlements · per-capability kill switches.
- Cost — every query is charged (minimum 10 MB); default cap 1 GiB per query. The screen shows the estimate; the amount charged is the one on the invoice.
- Limits — see the table above.
- Security — per-workspace scope, redaction, production guard, masked PII, named parameters.
- How to validate — in the console, when building a query, check the cost/bytes estimate shown before running it; a query that would exceed the billable-bytes cap is aborted with a warning. An administrator can check the environment's deep health at
GET /api/health?deep=1. - In short — query protection (bytes cap, timeout), cost estimate before running and per-workspace isolation, on a platform with a 99.5% SLA from Growth.