lavalake documentation

What the platform is, how it is built, and how to get it running.

• WelcomeWhat lavalake is, what it consists of, and who it is built for: a self-hosted lakehouse with governance, search and an AI assistant in a single installation.
• ArchitectureThe services of a lavalake installation and how they interact: gateway, console, agent, Keycloak, Trino, Iceberg, Nessie, MinIO, PostgreSQL and the local embedding model.
• Quick startThe whole platform on one machine in minutes: Docker Compose, one script, one address. No Kubernetes, no cloud key.
• The consoleA tour of the console's working areas: workspace, data, AI and semantics, operations and administration — and what each menu item is for.

The decisions behind it: authorization, search and the agent.

• Governance & identityDelegated identity, row and column policies, the WORM audit log and data lineage: how lavalake governs access and makes it provable.
• Semantics & searchMetrics and semantic views, plus vector, keyword and hybrid search with a local embedding model — and why policies apply to results too.
• The AI assistantHow the built-in agent works: delegated rights, curated tools behind a gate–audit–run chain, confirmation prompts for changes, and the platform's own MCP server.

Tasks, step by step.

• Importing dataThe five routes into the lakehouse: CSV and files, database connectors, documents, MCP ingestion and the fileshare crawler — with an assistant instead of a config file.
• SQL queriesThe worksheet over Trino: finding objects, querying Iceberg tables, travelling back in time, and creating views that BI tools can consume.
• Configuring the AI backendChoosing the language model behind the chat: a provider key, a self-hosted model, or per-user Claude sign-in — and what each means for your data.

Connecting other tools to the lakehouse.

• Tableau & BI toolsThe read-only HTTP feed for Tableau's REST API Connector: switching it on, authenticating, building the data source URL — under the analyst's identity, fully audited.
• HTTP API of the BI feedEndpoints, parameters, response shape and error codes of the read-only data feed — with examples for curl, Python, C# and PowerShell.

Install, configure, operate.

• Installing on KubernetesThe production path: the umbrella Helm chart, the three scripts for cluster, realm and proof — and the issuer lockstep that makes tokens line up.
• ConfigurationThe agent service's environment variables: required values, security switches, embeddings, audit archival — and what happens when one is missing.