Platform · Anthropic
Frontier reasoning, inside your governed data.
Anthropic's Claude models bring frontier reasoning to enterprise work, and through Snowflake Cortex they run inside your security perimeter, grounded in your governed data, with nothing shipped to an external endpoint.
Different ways of working, not a longer feature list.
Claude is available in Snowflake Cortex. Prompts and data stay under the same governance as every other query: there is no egress to a third-party endpoint to risk-assess and monitor.
Grounded in your models and Horizon Context, not the open internet. The answer is built from the same trusted definitions your reporting uses.
Anthropic's safety and alignment posture, enterprise controls and data-use commitments are built to survive a procurement and risk review, not just a demo.
Claude powers Cortex Agents and CoWork, and the Claude Agent SDK and Model Context Protocol let us build custom agents on your own guardrails.
Claude: the reasoning engine behind the agents.
Anthropic builds Claude, the frontier models that do the judgement-heavy work inside agentic analytics. On this platform they run where the data is governed, not on an endpoint your data has to travel to.
Claude (Sonnet & Opus)
Frontier reasoning for enterprise work
Long-context reasoning, dependable tool use and structured, reviewable output: the properties that make an agent reliable rather than a demo. Sonnet for volume, Opus for the hard problems.
- Long-context reasoning over enterprise documents and data
- Dependable tool use for multi-step agents
- Structured output a human can review
- Available natively in Snowflake Cortex
Claude in Snowflake Cortex
Claude where the data is governed
Call Claude from Cortex AISQL, Cortex Analyst and Cortex Agents. Prompts and data stay inside Snowflake's perimeter, under the same RBAC and masking as any other workload.
- No data egress to an external endpoint
- Same governance as every other query
- Powers Cortex Agents and CoWork
- No separate AI vendor to onboard to deploy
Claude Code & the Agent SDK
Build the agents, not just the prompts
An agentic coding assistant for data and analytics engineering, plus the Claude Agent SDK and Model Context Protocol (MCP) for connecting Claude to your governed tools and data.
- Accelerates dbt, pipeline and SQL work
- MCP connects Claude to governed systems
- Custom agents built on your own guardrails
- The same models in the IDE and in production
Not the benchmark scores. The reasons Claude on governed data pays back.
Transformation
AI stops being a side experiment and becomes part of how work gets done, grounded in your data, not a generic assistant bolted on the side.
Business process improvement
Claude handles the reasoning-heavy steps inside processes that used to wait on a person being free: summarising, classifying, drafting, checking.
Optimisation
Running Claude through Snowflake Cortex means one governed compute and security model for AI, not a parallel stack to run, secure and reconcile.
Cost saving
Model choice by task: Sonnet for volume, Opus for the hard problems, and no data-movement or separate inference infrastructure to pay for.
Future-proofing
Frontier capability improves under you. As the models get stronger, the agents you built on governed data get better without a re-platform.
Datalab is building its Anthropic partnership.
We are formalising our partnership with Anthropic. In the meantime we already run Claude in production the low-friction way, through Snowflake Cortex, and build custom agents on the Claude Agent SDK and MCP.
Governed deployment
Claude deployed via Snowflake Cortex: governed, no data egress, nothing new to onboard to get started.
Custom agents
Built with the Claude Agent SDK and Model Context Protocol, on your own guardrails, not a generic chatbot wrapper.
Risk review built in
Safety, data-use and governance review run with your risk team from the start, not requested after a pilot has already stalled.
AI on a governed foundation
Think, Build, Grow: AI capability built on a governed data foundation, not deployed ahead of one.
