dbt Semantic Layer
Read your existing semantic models. The Lanikaia metrics layer publishes back into the same surface.
01 / Platform
Built on Apache Polaris and Apache Iceberg. Governed by your IdP. Clear about what is beta.
Lanikaia is not a black box on top of your data. The catalog is Polaris, the storage is Iceberg, the identity is yours, and every artifact is a versioned snapshot you can read with the engines you already trust.
02 / Architecture
Four runtime surfaces, one catalog of record.
Apache Polaris (Iceberg REST catalog)
All datasets are catalog entries in Polaris. Projects map to namespaces. Grants and visibility are enforced at the catalog level, before queries run.
Apache Iceberg storage
Tables are Iceberg snapshots on object storage. Schema evolution, partition evolution, and time travel are first-class. Reads stay isolated from concurrent writes.
Conversational compute layer
The agent generates Python that runs in a sandboxed worker against Polaris-managed Iceberg tables. Output is a new snapshot plus the auditable code that produced it.
Workflow runtime
Promoted chats run on schedule labels inside the same execution surface as interactive sessions. Backfills, retries, and SLA monitoring are part of the Enterprise roadmap.
03 / Governance
Catalog-scoped permissions, IdP-rooted identity, queryable audit.
Identity flows from your IdP
SSO via OIDC, Google Workspace, and Microsoft Entra ID (Azure AD). Users keep their corporate identity. Group membership in your IdP is the source of truth for who can read which namespace.
Grants are catalog-scoped
Read, write, and own permissions live in Polaris. A project is a namespace; a chat references upstream namespaces; queries that touch tables a user cannot read are rejected before compute runs.
Every action is logged
Sign-ins, grant changes, chat invocations, table writes, API requests. The audit log is queryable like any other Iceberg table; one year later you can reconstruct who did what.
Lineage is the code you wrote
Lineage is not reconstructed from logs. Every commit captures the Python that produced the snapshot, the upstream snapshots it referenced, and the user identity that ran it.
04 / Surfaces
Four ways to consume what Lanikaia produces.
- APIVersioned HTTPPromote a dataset or function to a stable HTTP endpoint. Pinned to a Polaris namespace and a Lanikaia commit; clients consume schema and data without authoring SQL.
- MCPModel Context Protocol serverExpose the same artifacts as MCP tools so other agents (Claude, Cursor, in-house copilots) can read and act on them with native typing and authorization.
- BITableau / Looker / Power BIIceberg via Trino or Spark connectors. Existing BI seats continue to work; the Lanikaia commit they read is pinned by the catalog reference.
- Exceloutput-excel layerOutput any commit as an Excel file. Useful for finance and operations users who do not log into the platform. Google Sheets and bidirectional add-ins are on the roadmap.
05 / Integrations
Read from where the data is. Write to where the team works.
SAP / BusinessObjects
Read from SAP universes and BO reports. Treat them as ingest-layer sources without rewriting them.
Airflow / Dagster
Promoted Lanikaia workflows can be triggered from your existing scheduler, or run inside the Lanikaia runtime.
Trino / Spark
Execute queries on Lanikaia-managed Iceberg tables from your existing query engines. No data duplication.
Object storage
AWS S3, Google Cloud Storage, Azure Blob, MinIO. BYO bucket; Lanikaia stores Iceberg manifests and Parquet there.
Notebook / IDE
JupyterLab and VS Code read and write Lanikaia commits via the Iceberg client and the catalog REST endpoint.
06 / Security
What we do, and what stays under your control.
Lanikaia is designed for BYO bucket and BYO catalog. Iceberg snapshots and Parquet stay in your object storage; we never replicate them. The beta supports hosted evaluation paths and BYO deployments by agreement. Private link / VPC peering, SOC 2 audit, dedicated runtime tenancy, and contractual data-residency guarantees are Enterprise roadmap items.
- Data residencyBYO S3 / GCS / Azure Blob (planned)
- CatalogBYO Polaris, or hosted Polaris
- IdentityOIDC / Google / Microsoft Entra
- NetworkPublic; private link is on the Enterprise roadmap
- Audit logAll actions recorded; retention agreed by contract
- ComplianceBeta. SOC 2 audit in preparation for Enterprise GA
07 / Next