xAI's Grok models are now natively available on Databricks Agent Bricks, the announcement landing live at the Databricks 2026 Data + AI Summit on June 18, 2026. The integration lets enterprises build AI agents powered by Grok while keeping all of their data inside a governed Databricks Lakehouse environment.
Key Highlights
- Grok models are now natively callable inside Databricks Agent Bricks, Databricks' platform for building, deploying, and governing AI agents
- Enterprises can build agents that reason directly over both structured and unstructured data in their Lakehouse, without exporting data to an external pipeline
- All data handling stays under Unity Catalog governance, with zero data retention endpoints — model partners, including xAI, do not retain submitted data
- Databricks does not train foundation models on customer data
- Grok joins a model lineup that already includes OpenAI, Anthropic, Gemini, Qwen, and the newly added Kimi
What Agent Bricks Is
Agent Bricks is Databricks' developer platform for the full lifecycle of AI agents — building, deploying, and governing them. Since launching last year, the platform has grown to support more than 100,000 agents processing over one quadrillion tokens annually. Databricks frames the platform as solving "the missing 99%" — the infrastructure and governance work that surrounds the core agent loop.
The platform connects context, derived from data already sitting in the Lakehouse, with control and choice over which model powers each agent. That lets engineering teams build agents that operate on large volumes of enterprise data while staying inside a single governed environment.
The Grok Models
Two Grok variants are highlighted for the Databricks integration:
- Grok 4.3 — xAI's flagship reasoning model, featuring a 1-million-token context window. Priced at $1.25 per million input tokens and $2.50 per million output tokens.
- Grok Build 0.1 — a coding-focused variant, priced at $1.00 per million input tokens and $2.00 per million output tokens.
Governance and Data Control
The integration is built around Databricks' governance layer. Agents, tools, and models are registered into Unity Catalog for unified governance alongside data assets, and the Unity AI Gateway provides model discovery, access controls, cost monitoring with per-user and per-group budgets, and intelligent traffic routing.
Crucially for regulated industries, Databricks has confirmed that model partners do not retain data submitted through these features. The platform uses zero data retention endpoints, and Databricks itself does not train its foundation models on customer data.
Impact
The deal continues a steady march of Grok into enterprise infrastructure. Grok has progressively landed on Oracle Cloud Infrastructure, Microsoft Azure AI Foundry, and Amazon Bedrock, and now reaches the data-platform layer where enterprise data already lives. By meeting that data in place, the integration removes one of the biggest friction points in enterprise AI: moving sensitive data out of a governed environment to reach a capable model.
For Databricks, adding Grok widens the model menu inside Agent Bricks at a moment when enterprises increasingly want to switch providers without re-architecting. As Gregory Rokita, VP of Technology at Edmunds, put it: "Databricks gives us a secure, governed foundation to run multiple models and switch providers as our needs evolve. All while keeping costs in check."
MENA Angle
For organizations across the MENA region — where data residency, sovereignty, and sector regulations such as Saudi Arabia's PDPL shape every AI decision — a model that runs against data inside a governed Lakehouse, with no retention by the provider, lowers the compliance bar for adopting frontier AI. Enterprises in banking, energy, and government can evaluate Grok-powered agents without first solving the data-movement problem that has stalled many regional AI pilots.
What's Next
With Grok now alongside OpenAI, Anthropic, Gemini, Qwen, and Kimi inside Agent Bricks, the competitive story shifts from "which model" to "which governed platform." Expect enterprises to A/B test multiple frontier models on the same internal data and routing rules, choosing per workload rather than committing to a single vendor. The model layer is becoming a swappable component; the data and governance layer is where the lock-in now lives.
Source: Databricks