LoginSign Up
← Back to all sparks
K

Keboola

ANALYTICS
Velocity7.5

End-to-end data operations platform for pipelines, transformations and orchestration

keboola.com ↗

Keboola's Kai AI assistant hits GA, completing the pivot from data platform to AI-native pipeline orchestration layer.

data-platformai-agentsmcpetlsnowflake
◆Current state
Keboola's Kai AI assistant is now generally available for multi-tenant contracted customers, marking the formal transition from experimental feature to production-ready product. The platform is simultaneously navigating Snowflake's forced password authentication deprecation — a platform-level forcing function requiring customer migrations before September 7. Recent work also includes the MCP server gaining unified cross-project authentication, which enables AI coding assistants to act as first-class operators across a full Keboola stack.
◆Where it's heading
Keboola is building toward a model where AI agents can autonomously manage the data pipeline lifecycle. The MCP server's unified auth is a signal: the target is a world where a developer's coding assistant can browse, create, and modify Keboola pipelines without a human navigating the UI. Kai GA and the MCP expansion are the same thesis from two directions — AI as interface, not AI as feature. Branched storage expansion to BigQuery and Flows improvements in the background are the operational stability layer those agents will depend on.
◆Prediction
Kai gaining the ability to create and modify Flows, and the MCP server expanding its coverage to transformation and workspace management, are the two most visible next moves. The Branched Storage on BigQuery release is a prerequisite for Branches 2.0 approval workflows, which would give Kai a way to propose and commit pipeline changes with human review gates.

◆Recent moves

  1. 1d ago

    Kai is Now Generally Available (Multi-Tenant, Contracted Customers)

    ⚡ SPARK

    Kai reaching GA for contracted customers is not a label change — it's Keboola formally committing AI as the primary interaction layer for its platform, backed by production SLAs. In the trajectory of AI-native data tooling, this positions Keboola ahead of traditional ETL platforms that treat AI as an add-on rather than the interface.

    View source ↗
  2. 6d ago

    Migrate Snowflake Workspaces from Password to Key Pair Auth

    Snowflake's SERVICE user transition is breaking password-authenticated workspaces, and Keboola's migration guide gives customers the specific path to repair them without losing workspace content — a reactive but necessary operational update ahead of the September 7 forced migration.

    View source ↗
  3. 22d ago

    Keboola MCP: Log In Once, Work Across Every Project

    ⚡ SPARK

    The MCP server's one-time browser login, spanning all projects on a stack, removes the per-project token workflow that made AI-assisted pipeline management impractical at scale — this is the enablement layer that makes Kai's GA meaningful for developer-facing AI tooling.

    View source ↗
  4. 27d ago

    Python 3.10 Deprecation for Streamlit Apps (September 2026)

    A deprecation notice for Python 3.10 Streamlit apps — the default runtime switches to 3.11 on September 24, with the 3.10 image stopping security fixes. Operational housekeeping, not a capability change.

    View source ↗
  5. 1mo ago

    Flows: New Condition Operators and Run Selected Tasks

    Adding 'continue on failure' logic, per-phase conditions, and selective task execution to Flows gives data engineers meaningful control over failure handling without rewriting pipelines — the kind of operational control that's often a blocker for using visual pipeline builders in production.

    View source ↗
  6. 1mo ago

    Branched Storage on BigQuery

    Extending branched storage to BigQuery brings it to parity with Snowflake projects and is a stated prerequisite for Branches 2.0 — the upcoming approval workflow and schema change tracking feature that would give AI agents a governed path to propose pipeline changes.

    View source ↗