Skip to content
MCP server built inHow that works
Back to the platform overview

Roadmap

What lavalake does today and what comes next

This page collects what is scattered across the topic pages. Anything listed as available is built into the product and shipped in a release. Anything planned says planned – including where a competitor has had it for years.

AI and assistant

The layer this is about: questions to your own data, under the rights of whoever asks.

  • AvailableAssistant with curated tools, confirmation before every change
  • AvailableMCP server that passes through the user's own permissions
  • AvailableRegistry for third-party MCP servers in the console
  • AvailableChoice of language model per installation, self-hosted included
  • AvailableSigning in with your own Claude subscription as agent mode
  • AvailableDashboards fully editable through the assistant
  • PlannedNatural language to SQL against curated metrics
  • PlannedAI functions directly in SQL, as Trino functions
  • PlannedAWS Bedrock as a language-model provider

Lakehouse and queries

Open formats on your own storage, so the data stays readable without us.

  • AvailableApache Iceberg on S3-compatible object storage
  • AvailableVersioned catalog with time travel and branching
  • AvailableTrino as the query engine, SQL worksheet in the console
  • AvailableData browser across catalog, schema and table
  • PlannedElastic compute clusters with auto-suspend
  • PlannedIncremental materialization for expensive queries

Data ingestion

Four routes into the lakehouse, each with an assistant instead of a config file.

  • AvailableImport assistant with modelling and cleanup before loading
  • AvailableConnectors for PostgreSQL, MySQL and Microsoft SQL Server
  • AvailableFiles, documents and a file crawler as sources
  • AvailableThird-party MCP servers as a data source
  • PlannedContinuous ingestion instead of full load runs
  • PlannedChange data capture
  • PlannedScheduled tasks

Search, metrics and BI

One definition per metric – the same number in chat, dashboard and BI tool.

  • AvailableHybrid search combining vector and keyword retrieval
  • AvailableEmbeddings from a local model, no external service
  • AvailableCurated metrics and semantic views
  • AvailableDashboards that compute under each viewer's own rights
  • In progressTableau reads the lakehouse through a read-only feed
  • In progressPower BI through a bundled Power Query connector
  • PlannedAlerting and scheduled reports

Permissions and evidence

Nobody gets around the permissions – the agent included.

  • AvailableDelegated identity through Keycloak, no service account
  • AvailableRow and column policies in Trino
  • AvailableTamper-evident audit log as a hash chain
  • AvailableLineage and access history per table
  • AvailableUser and role administration in the console
  • PlannedRole hierarchy and ownership
  • PlannedAccess, deletion, retention and classification as features

Operations

One machine to get to know it, a Helm chart to run it. The rest is maturity.

  • AvailableCompose stack for a single machine, Helm chart for Kubernetes
  • AvailableMonitoring and audit inspection in the console
  • PlannedHigh availability without single points of failure
  • PlannedMetrics, logs, traces, alerting and SLOs
  • PlannedTLS between all services
  • PlannedSecrets through Vault or External Secrets
  • PlannedBackup and disaster recovery
  • PlannedAir-gapped operation with no outbound connections
  • PlannedBudgets and quotas for compute cost

Deliberately not planned

Large product areas we are not taking on. Anyone who needs them is better served by a cloud warehouse – saying so is more honest than leaving it open.

  • Not plannedData marketplace and listings
  • Not plannedSecure data sharing between organisations
  • Not plannedReader accounts for external parties without their own installation
  • Not plannedCross-region replication
  • Not plannedRunning application code inside the database
  • Not plannedHosting third-party applications on the platform

Something missing, or something urgent?

The order comes from customer questions, not from a drawing board. Tell us what you need – that moves the list more than a competitive comparison does.

Arrange a conversation