CLAIM
Governed analytics. Right-sized.
Governed SQL over your object storage,without carrying a larger platform than the workload requires.
Clarté unifies SQL execution, policy, scoped storage authority, audit, and Kubernetes operations in one customer-controlled platform.
BUILT ON DUCKDB · DUCKLAKE · POSTGRESQL · KUBERNETES · S3-COMPATIBLE STORAGE
$ SELECT region, sum(amount) FROM orders GROUP BY region;
- identity
- analyst@eu-team · oidc verified
- policy
- eu_customer_access · compiled
- rows
- region IN ('EU-WEST', 'EU-CENTRAL')
- columns
- email masked · ssn denied
- scope
- s3://lake/orders/region=EU-*/
- audit
- decision recorded
✓ governed result returned
IDENTITY→POLICY→SQL→STORAGE→AUDIT
01 / Why this platform exists
Size the platform for the query, not the archive.
The volume you retain is not the volume a typical query reads. Most analytical queries touch recent periods, selected columns, and bounded partitions — yet platforms keep being sized for the archive.
Retained dataWHAT YOU STORE
Query working setWHAT A TYPICAL QUERY READS
Five measurements decide the architecture:
- 01Bytes readWhat are the median and 95th-percentile bytes read per query?
- 02ShufflesHow often does a query genuinely require a distributed shuffle?
- 03ConcurrencyWhat is the real peak concurrent-query demand?
- 04LatencyWhat response time does the business actually require?
- 05EffortHow much engineering time goes into operating the platform?
When the measured answers do not justify a distributed engine, Clarté is the right-sized platform — with governance intact.Check your workload shape ↗
The platform
Everything governed analytics needs. Nothing it doesn't.
Clarté is a complete platform with a deliberate boundary: one governed path from sign-in to storage, operated on your infrastructure.
THE PLATFORM BOUNDARY INCLUDES
- Governed analytical SQL on DuckDB and DuckLake
- External OIDC identity and policy-driven access control
- Short-lived scoped credentials for S3-compatible object storage
- Query-to-audit correlation and durable audit records
- Kubernetes-native operations through operator and control plane
- PostgreSQL for metadata and control-plane state
- Your object storage as the analytical data boundary
FIVE GOVERNED SURFACES IN THE BROWSER
- 01SQLWrite and run governed analytical queries.
- 02CatalogDiscover datasets and the access attached to them.
- 03QueriesExecution history, status, and runtime context.
- 04AuditIdentity, policy decision, and storage scope in one record.
- 05AdminPlatform health, governance state, and configuration.
One authorization chain
A policy in the BI layer means nothing if storage disagrees.
Identity, SQL policy, storage permissions, and audit usually live in separate systems that drift apart. Clarté treats them as one chain, evaluated per session.
- 01IdentityExternal OIDCYour identity provider authenticates every session. Groups and claims map to Clarté roles — no separate user database.
- 02PolicyResolved before SQLRow filters, column masks, and dataset grants are compiled for the session before a single statement runs.
- 03SQLDuckDB + DuckLakeOne deliberate analytical runtime executes governed queries through policy-shaped views.
- 04StorageScoped credentialsEach session holds a short-lived credential that reaches only permitted prefixes — enforced by your object store, not by a string check.
- 05AuditCorrelated recordsEvery attempt becomes a durable record tied to identity, policy version, and the exact credential used.
IDENTITY → POLICY → SQL → STORAGE → AUDIT — evaluated per session, recorded per query
Customer-controlled
Runs in your cluster. Reads from your storage. Trusts your identity provider.
Clarté ships as software into a Kubernetes environment you approve — European cloud or self-hosted. There is no vendor-hosted control plane and analytical data never leaves your object storage.
No vendor-hosted control plane
The operator, control plane, and query surface run inside your cluster.
Storage authority stays scoped
Sessions hold short-lived credentials limited to permitted prefixes.
Identity stays yours
Your OIDC issuer authenticates every session; claims map to roles.
Evidence
Proof you can inspect, not adjectives.
Clarté's claims link to published, reproducible evidence — each entry states what was claimed, what was exercised, and where to inspect it.
See the evidence
06 / A deliberate fit
Built for bounded workloads. Honest when yours isn't.
Clarté is narrow by design. If your workload needs a distributed engine, the evaluation will say so — a not-fit answer is a valid outcome, stated plainly.
Evaluate a real workloadSTRONG FIT WHEN
- Typical queries read bounded partitions, recent periods, or selected columns.
- Different roles need different row, column, or dataset access.
- Kubernetes and object storage are already part of your operating model.
- Deployment control matters as much as query performance.
USE A DIFFERENT ARCHITECTURE WHEN
- Petabyte-wide scans are routine, not exceptional.
- Large distributed joins and shuffles dominate the workload.
- Thousands of simultaneous interactive queries are required.
- Sub-second streaming is the primary requirement.
07 / Workload Fit Pilot
Ten business days to a defensible decision.
A fixed-scope, paid engagement that tests one real workload against Clarté and produces the evidence for an adoption, deployment-scope, or no-go decision.
FITCONDITIONAL FITNOT FIT
Explore the pilot- 1
- representative dataset or approved workload slice
- 3
- representative queries, at most
- 2
- identity roles with different access
- 1
- row, column, or dataset-access policy
- 1
- agreed deployment environment
- 10
- business days once prerequisites are ready