TechPrufer
Get started
Hex
Hex logoHex · Feature trail

AI Gateway

6 posts · Nov 2024 – Jul 2026

What this trail covers

Hex has evolved its AI capabilities by integrating with Snowflake Cortex, allowing users to leverage Claude models directly within their Snowflake instances. This integration enhances control, governance, and observability for AI workloads, keeping data within the customer's Snowflake account. Users can already call Cortex functions in SQL and add Cortex APIs to Python cells in Hex, and soon will be able to back Hex's Magic tools with Claude 3.5 on Cortex. This post details the concept of an AI Gateway and the benefits of using Claude 3.5 on Cortex for Hex's Magic tools, emphasizing data residency and enhanced AI workload management.

All Hex trails
  1. What Is an AI Gateway and Why Should You Care?This post introduces and explains the concept of an AI gateway as a critical infrastructure component for managing LLM traffic. It details the functionalities of an AI gateway, including request routing, token/dollar-based rate limiting, credential management, prompt caching, failover capabilities, and cost attribution. The post contrasts AI gateways with traditional API gateways and outlines scenarios where an AI gateway is necessary, such as in multi-model environments, when multiple teams share LLM access, or when AI is on the critical path. It also discusses the integration of analytics platforms with AI gateways, advocating for BYOK patterns. The post aligns with the existing 'AI Gateway' feature thread by elaborating on the architectural and operational aspects of managing AI model interactions.
  2. AI Sprawl: Causes, Risks & 7-Step Control PlanThis post details a 7-step plan to control AI sprawl, focusing on architectural solutions rather than policy. Key steps include inventorying existing AI tools, centralizing metadata into a shared layer (semantic layer), defining high-value metrics, and establishing a governed context layer. It emphasizes that shared context, where AI systems draw from a single, governed layer of business definitions, is crucial for controlling AI sprawl and ensuring metric trust. The plan advocates for an incremental approach, starting with metadata consolidation and then defining key metrics, rather than a big-bang migration. It highlights the risks of AI sprawl, such as metric trust collapse and a hidden tax on data teams, and proposes solutions like semantic model sync to connect governed measures to AI tools.
  3. Shadow AI governance: strategies to rein in rogue modelsThis post details strategies for governing 'shadow AI' (unauthorized AI tool usage), emphasizing the risks of data exposure and metric inconsistency. It advocates for building governance into platform architecture, implementing risk-based approval processes, aligning with regulatory frameworks (NIST AI Risk Management Framework, EU AI Act), providing approved alternatives, and creating visibility through monitoring tools. The core argument is that making governed analytics faster and easier than alternatives is the best defense against shadow AI, highlighting Hex's AI-assisted platform and semantic modeling as solutions.
  4. MCPs: What they are and how you can use them with Hex's APIThis post introduces the concept of Model Context Protocols (MCPs) and demonstrates how to build an MCP server that interfaces with Hex's public API. It details the steps for installing and configuring the `hex-mcp` package, enabling AI assistants like Cursor and Claude to search for, trigger, and check the status of Hex projects. This enhances the developer workflow by allowing interaction with Hex projects from within AI-assisted development environments, improving focus and streamlining data transformation and job orchestration.
  5. hex cortex overview, all your questions answeredThis post details the integration of Hex with Snowflake Cortex, focusing on the advantages of using Snowpark and Cortex on Hex over standalone Snowflake SQL/Python/Cortex. It highlights Hex's 'Snowpark Easy Button' for secure session creation, seamless Snowpark dataframe integration in SQL cells, and Snowpark-aware visualizations that push aggregation to Snowflake. The post also addresses questions about swapping models (OpenAI vs. Cortex), evaluating model suitability, main considerations for Cortex functions (availability, cost, context window), fine-tuning models with Snowflake Registry, handling large datasets with pushdown compute, and using open-source models in Snowpark Container Services versus calling them via `complete()` function.
  7. Announcing: Magic support for Claude on Snowflake CortexThis post announces the upcoming integration of Hex's Magic tools with Claude 3.5 models running on Snowflake Cortex. It highlights that users can already leverage Cortex for SQL functions and Python cells within Hex. The key evolution is the planned ability for Hex's Magic tools (Ask Magic, etc.) to utilize Claude 3.5 on Cortex, providing enhanced control, governance, and observability for AI workloads while keeping data within the customer's Snowflake account. It also mentions the existing integration with Cortex for SQL and Python cells.