Data sources · The foundation

Connect once. Query forever.

Arcus starts warehouse-first with BigQuery, Snowflake, and Postgres. When the metric only lives in a source app, scoped Pipedream connections can provide first-party data, and Copass keeps the source context, snapshots, and evidence trail together.

arcus.ai/data/sources
Data sources
3 warehousesScoped app sources+ New
Core 3
Warehouses 3
App accounts Scoped
Memory Copass
Snowflake
BigQuery
Postgres
Additional warehouses by request
Copass context
Stripe via Pipedream or warehouse
Salesforce via Pipedream or warehouse
HubSpot via Pipedream or warehouse
GA4 via Pipedream or warehouse
Meta Ads via Pipedream or warehouse
Google Ads via Pipedream or warehouse
More by request
3
Core warehouses: BigQuery, Snowflake, and Postgres.
Read
Connections are scoped to read-only warehouse roles.
Fresh
Freshness follows the source: warehouse refreshes or source-native pulls.
0
SQL when available. Source-native APIs when that is the real first-party system.
01 · Semantic layer

Speaks your business, not your schema.

Most tools ask you to translate "active customer" into a table join every time. Arcus learns it once. Define active_customer, paid_cohort, north_star — Arcus uses your definitions, every time.

  • Configurable by tenant. Arcus uses the business definitions approved for each workspace.
  • Reviewable changes. Metric definitions should have owners, notes, and a change trail.
  • Existing models welcome. dbt and BI metadata can inform setup when available, but launch scope is reviewed case by case.
semantic.yml · v23
# How "active customer" is calculated, everywhere. metric: active_customer label: "Active customer" source: stripe.charges filter: last_charge < 90d AND status = 'paid' owner: @finance
# Used in 47 threads, 12 dashboards, 3 alerts metric: north_star label: "Weekly Active Buyers" expr: count(distinct user) where activated cadence: weekly
02 · Lineage + freshness

Trace every number to its source row.

"Where did this number come from?" Click. See the chain Arcus used: answer, metric, SQL or source request, warehouse table or connected account, Copass context, and available source evidence. Freshness checks depend on the fields and metadata connected during setup.

  • Answer lineage. Across warehouse queries, source-native pulls, Copass memory, and answer artifacts.
  • Freshness checks. Set expectations per source and flag stale warehouse tables.
  • Schema drift review. Column changes can be flagged when metadata coverage supports it.
freshness · last 24h
stripe.charges
42s ago · streaming
meta_ads_daily
11m ago · hourly
shopify.orders
3h late · investigating
salesforce.opportunities
22m ago · hourly
ga4.events
1m ago · streaming
lineage · north_star
stripe.chargesdim_customeractive_customernorth_star
03 · Governance

Row-level access. Audit-ready.

Arcus is designed around tenant isolation, read-only warehouse access, and application-level permission checks. Row-level and role-specific policies are scoped during setup and must match the customer's warehouse model.

  • Warehouse-aware permissions. Launch setup uses read-only roles and tenant-scoped application access.
  • Row-level rules. Scoped during pilot; not assumed as public self-serve.
  • Audit log. Query and access history are captured for review; SIEM export is contract scope.
GDPRAudit trailTenant isolationRBAC
access · stripe.charges
JK
Finance owner
finance@client.example · Finance
adminall rows
MO
Growth owner
growth@client.example · Growth
analystaggregates only
DT
CS owner
cs@client.example · CS
viewertier=enterprise
RA
Engineering owner
eng@client.example · Eng
viewerno PII
Audit log retained by agreement — query and access history are captured for review; export options are scoped during security review.
04 · Scoped extensions

When the source is outside the warehouse, route it.

Pipedream, Copass ingestion, file drops, and custom APIs are scoped by tenant, source, auth model, evidence needs, and security review. Arcus should tell users which path is authoritative for each answer.

REST API

Source-native pulls when needed.

Pipedream-backed app accounts can be authoritative first-party data when the tenant has no warehouse table for that metric.

CSV / Excel / Sheets

Evidence snapshots into memory.

Useful source summaries, onboarding docs, and citations can be ingested into Copass so Arcus has durable context without overriding SQL or source-native facts.

Custom SQL

Write a query. Save it as context.

Sometimes raw SQL is the answer. Save reviewed queries as reusable context when that fits the tenant's semantic layer and evidence model.

The boring layer that makes everything else magic.

Connect a warehouse, scoped app account, or Copass source. Use guided setup when source coverage, permissions, citations, or semantics need review.