Skip to content
Roadmap

What's GA, what's coming.

Every Databricks product and feature we track, with the status it was last seen in. Statuses come from Databricks docs release notes, or from a public Databricks announcement where no note exists yet. A claim from an announcement is checked by hand before it changes a status. Independently maintained, not affiliated with Databricks. Last updated today.

180 tracked·108 GA·59 in preview·11 announced

AI & ML

Agent Bricks, Genie, MLflow, AI Search.

22
  • Agent BricksGAThe Databricks agent developer platform for building, deploying and governing AI agents in code, with any framework and any model. Its first no-code agents, Knowledge Assistant and Supervisor Agent, reached GA in early 2026 and became legacy in October 2026. Newer parts such as the Agent Bricks CLI, Databricks Sandbox and managed agent memory are in Beta.3
  • AI Functions REST APIGAA REST API for calling ai_parse_document, ai_extract and ai_classify in real time from applications and agents, for example as soon as a document arrives. Databricks also markets it as the Document Intelligence real-time REST APIs.
  • AI Gateway (legacy) for serving endpointsDeprecatedThe original Mosaic AI Gateway for model serving endpoints, which adds permission and rate limiting, payload logging, usage tracking, AI guardrails, fallbacks and traffic splitting to individual endpoints. Databricks now calls it AI Gateway (legacy) and recommends the newer Unity Gateway instead.
  • AI RuntimeGAServerless NVIDIA GPU compute inside Databricks for training and fine-tuning models next to governed data, with GPUs that start in seconds and no GPU infrastructure to manage. It was first released as serverless GPU compute (Beta, June 2025), was named AI Runtime when it entered Public Preview in March 2026, and Databricks announced general availability on October 21, 2026. Since August 27, 2026, Genie Code also works in notebooks connected to AI Runtime, where it generates distributed training code, resolves environment and dependency issues, investigates GPU and distributed workload failures, and migrates Foundation Model Fine-tuning API and classic Databricks Runtime ML GPU notebooks; the release note placed this in Public Preview and no later note has moved it, although the docs page shows no preview label today.
  • AI Search (formerly Vector Search)GABuilds search indexes from Unity Catalog tables (Delta tables, streaming tables, materialized views and managed Iceberg v3 tables) and serves vector similarity, hybrid and full-text keyword queries from AI Search endpoints through a REST API or Python SDK; full-text search is in Beta. Delta Sync indexes stay in sync with their source table, with embeddings computed by Databricks or supplied by the user, while Direct Vector Access indexes are updated through the API. Standard endpoints hold up to 320 million vectors at dimension 768, and storage-optimized endpoints over one billion.7
  • ai_decideBetaA SQL AI function that answers one or more questions about text or structured data with a probability, a choice from named criteria, or a score on a scale. It is built for routing, triage and prioritization over governed data and can also be called through a REST API for real-time apps and agents. Since October 27, 2026 it can also be used in Lakeflow Designer's AI Function operator (Beta).
  • ai_enrichBetaA SQL AI function that generates new columns for each row from a schema you define, filling them from the input content and, when asked, grounding them in knowledge sources such as AI Search indexes, web search or a search tool on an external MCP server registered in Unity Gateway. It is in Beta behind the Ai Enrich workspace preview; October 20, 2026, when MCP grounding arrived (Beta), is the first dated release note that names it with its tier.
  • ai_extract precision modeGAA mode of the ai_extract function for complex document extraction, such as multi-page documents, schemas with 50 or more fields, many extracted items per document and conditional field logic. It pairs custom fine-tuned extraction models with an agentic harness that extracts in parallel with subagents and merges the results, and it is turned on with the precision mode setting in ai_extract or a toggle in the Information Extraction UI. It was previously listed here as Document Intelligence Precision Mode, after the title of the launch blog.
  • Claude Fable 5 on DatabricksGAAnthropic's Claude Fable 5 model, offered as a Databricks-hosted model through Foundation Model APIs pay-per-token on AWS, Azure and Google Cloud, with access governed through Unity Gateway.
  • CustomerLakeBetaDatabricks' agentic customer data platform (CDP) that runs inside Databricks under Unity Catalog governance, giving marketers Customer 360 profiles, natural language audience building with Genie, Campaign Agents and activation to martech and adtech tools without copying data out. It launched in Private Preview at Data + AI Summit on June 16, 2026 and reached Beta on October 19, 2026 when Campaign Agents moved to Beta, while Profile Agents are in Private Preview; access is through the Databricks account team.1
  • Databricks Feature StoreGAThe Databricks central registry for the features used in AI and ML models, governed in Unity Catalog with lineage, point-in-time joins, and feature sharing and discovery across the organization. Teams author features as Feature Views or feature tables and can publish them to Online Feature Stores, powered by Lakebase, for real-time serving. The older Workspace Feature Store for workspaces without Unity Catalog is deprecated.1
  • Databricks GenieGAThe Databricks family of AI experiences for working with data in natural language. It includes Genie One (formerly Databricks One) for business users, Genie Agents (formerly Genie spaces) that data teams curate, Genie Code (formerly Databricks Assistant) for developers, and the Databricks Genie mobile, Slack and Microsoft Teams apps. Genie spaces became generally available in December 2024, while the mobile, Slack and Teams apps are still in Public Preview.9
  • Foundation Model APIs provisioned throughputGAA Foundation Model APIs serving mode that runs foundation models on dedicated capacity with performance guarantees, and the mode Databricks recommends for production workloads. It serves base models, fine-tuned variants and fully custom weights of supported model architectures, with Databricks managing the serving infrastructure.
  • Genesis Workbench (solution accelerator)GAAn open-source Databricks solution accelerator for drug discovery and genomics that runs NVIDIA BioNeMo, Parabricks and other biology models behind a no-code app inside your own workspace. Anyone can deploy it from GitHub, but it is a Databricks Industry Solutions blueprint, not a platform feature with a preview or GA label.
  • Genie OntologyPublic previewThe context layer that gives Genie One and Genie Code a shared map of an organization's business knowledge. It combines Unity Catalog semantics, such as metric views, with snippets that Genie extracts and ranks from existing assets such as dashboards, SQL queries and Genie Agents, and it uses only snippets from assets the user has permission to see.1
  • MLflowGAMLflow is an open-source platform for managing the end-to-end machine learning lifecycle.2
  • Mosaic AI (retired brand)DeprecatedFormer umbrella brand for Databricks AI and ML tools such as Model Serving, Vector Search, Agent Framework and Agent Evaluation. Databricks has dropped the name: Agent Bricks became the umbrella for all its AI capabilities in March 2026, and the tools continue under plain names.
  • OmnigentBetaOpen-source meta-harness for composing, controlling, and sharing multiple AI agents and agent harnesses, also offered as a managed Databricks version.
  • OpenAI GPT-5.5GAOpenAI's GPT-5.5 and GPT-5.5 Pro models are available as Databricks-hosted models in Model Serving, called through Foundation Model APIs pay-per-token.
  • PipelineIQ (Databricks internal tool)AnnouncedPipelineIQ is a sales pipeline tool that Databricks built for its own field sales organization on Foundation Model APIs, Unity Catalog, Delta Lake and AI/BI dashboards. It gives each sales opportunity a Walk, Pivot or Accelerate recommendation with a rationale and an action plan. Databricks shared it in a blog as an example of using its own platform; it is not a product or feature that customers can turn on.
  • Predictive AI FunctionsBetaA family of SQL functions for forecasting and analysis: ai_forecast version 2 projects time series forward with a time series foundation model, ai_top_drivers ranks the dimension values that contribute most to a change in a metric, and ai_detect_anomalies, ai_predict_value and ai_predict_class detect anomalies and predict values and classes. A workspace admin turns them on with the Predictive AI Functions option on the Previews page. Genie can call ai_top_drivers, and Databricks says the other functions are coming to Genie.
  • WASM Postgres for AI Agent SandboxesAnnouncedEmbeds lightweight WASM-based Postgres databases (PGlite) directly inside AI agent sandboxes with real-time synchronization back to Lakebase.

How this list is maintained

Every entry links to a public Databricks source. Each night we read the Databricks blog and the docs release-note feed. A status moves forward on its own only when a release note states it; a claim from a blog post waits for a check by hand before it changes anything. We do not see the private roadmap Databricks shares with customers, so a feature that has not been mentioned in public is not here. See a stale entry or a missing one? Tell us. See what changed.