Your revenue opssupportBImarketingcustomer successfinancesalescomplianceon‑callanalytics
agents need
a data platform.

✓ Dinobase semantic layer

stripe.subscriptions — "Active and historical subscriptions"

  customer_id  VARCHAR  → stripe.customers.id  [FK]
  plan_id      VARCHAR  → stripe.plans.id       [FK]
  status       VARCHAR  "active|past_due|canceled|trialing"
  created      INTEGER  "Unix timestamp — use to_timestamp()"
  amount       INTEGER  "Amount in cents"

Related: stripe.customers, stripe.invoices

Agent knows the join path, the enum values, the timestamp format, and which fields are PII — before writing a single query.

✓ SQL mutations

dinobase query "UPDATE stripe.customers
  SET name = 'Acme Inc'
  WHERE id = 'cus_123'"
# → mutation_id: mut_abc123
# → rows_affected: 1
# → side_effects: will call Stripe API

dinobase confirm mut_abc123
# ✓ Stripe API called
# ✓ Local data updated

One SQL primitive for every source. Preview before anything executes. Destructive DDL (DROP, ALTER, TRUNCATE) blocked entirely.

✓ Python execution layer

# Agents call exec_code(...) with a code snippet
from dinobase.query.engine import QueryEngine
from dinobase.db import DinobaseDB
from dinobase.mcp import call

with DinobaseDB() as db:
    rows = QueryEngine(db).execute("""
        SELECT email, mrr FROM stripe.customers
        WHERE status = 'past_due'
        ORDER BY mrr DESC LIMIT 10
    """)["rows"]

for row in rows:
    t = call("zendesk.tickets_for_user", email=row["email"])
    row["open_tickets"] = len(t.get("open", []))

result = sorted(rows, key=lambda r: -r["open_tickets"])

SQL + MCP + Python — one tool call, zero ping-pong. Your agent writes the orchestration; Dinobase runs it.