January 2025 Summaries

The Platform Dashboard for Hasura Data Delivery Network (DDN) is a comprehensive solution to elevate operational efficiency, data observability, and decision-making in modern application development. It provides a centralized interface with critical insights into Hasura DDN operations, empowering platform owners and developers to optimize API operations, ensure compliance and security, and streamline operations. The dashboard offers four key sections: Security and governance, API reliability, API usage trends, and Developer productivity, each designed to address critical aspects of observability and control. It is now available to all Hasura DDN users on Base plans and up, and can be accessed by logging in to the console and navigating to the Platform Dashboard section.

Build Safer AI Assistants with PromptQL and Human-in-the-Loop Guardrails` In today's AI landscape, guardrails are essential safeguards that protect both businesses and users from errors and unintended consequences. One of the most effective guardrails is human-in-the-loop (HITL) oversight, which combines computational power with discernment and accountability. This blog explores why HITL systems are crucial, how they align with agentic AI principles, and how PromptQL simplifies their implementation in AI Assistant interfaces. With PromptQL, developers can create and run query plans that enable agentic AI workflows to be both intelligent and controllable. The system's modular design simplifies HITL implementation, making it easier to build safer and smarter AI workflows. By integrating human-in-the-loop oversight, we can find the right balance between autonomy and accountability, ensuring that AI systems are not only efficient but also safe and reliable.

The text discusses the evolution of Hasura Cloud's monitoring system, which has become increasingly complex due to the growth of thousands of projects relying on its powerful GraphQL engine. The existing system relies on TimescaleDB and log ingestion, but it faces scalability issues such as high I/O speeds, delayed aggregate metrics, and frequent disk space issues. To address these challenges, Hasura introduces a new observability architecture that leverages Grafana, Prometheus, and its Data Delivery Network (DDN) to provide better insight into project performance, simplify troubleshooting, and offer granular, project-specific observability. The new system integrates the Wild GraphQL Data Source plugin with DDN to enable secure and scalable data delivery, allowing users to monitor their projects more easily and efficiently.

PromptQL has introduced several updates to enhance its power, accessibility, and collaboration capabilities, focusing on simplifying workflows. The platform now offers pre-configured LLM keys, making it easier for users to get started without the hassle of setting up their own API key. Additionally, Hasura's built-in LLM key uses the same backend as recommended models from providers like Anthropic or OpenAI. PromptQL has also launched a new Program API Endpoint, allowing developers to trigger and execute AI-driven query planning and execution within their existing workflows and external systems. This feature enables automation of complex tasks through Python-based scripts generated by LLMs. Furthermore, the platform now features a Public PromptQL Playground, making projects publicly accessible, and a new feature called Public Thread Sharing, which allows users to share specific chat threads with anyone without requiring login or signup. The platform also introduces Saved and Shareable Prompts, streamlining collaboration and user onboarding.

The PromptQL Program API offers dynamic integrations with data sources through its Python functions, allowing AI models to access and interact with all data under a unified schema. With the introduction of this API, developers can now trigger PromptQL Programs via HTTP, bringing flexibility and intelligence to business workflows. PromptQL Programs are generated by LLMs and enable read/write/search capabilities, interacting with databases, external APIs, and utilizing LLM's capabilities such as classification, summarization, and extraction. The Program API uses an ephemeral Hasura DDN token for authentication by default, but production environments require robust authentication mechanisms.

Hidden patterns in data can often go undetected by traditional validation approaches, but anomaly detection can reveal subtle clues that signal deeper issues. Anomaly detection at the egress point, where data flows to decision-makers, presents unique challenges and opportunities, including maximum business impact, rich context availability, and cross-system pattern recognition. Data composition capabilities have made these patterns more common and critical to detect, and modern low-data-movement tools provide unprecedented visibility into novel combinations. Building adaptive systems with anomaly detection as a continuous feedback mechanism can identify semantic inconsistencies, temporal anomalies, business rule violations, and data quality issues. The AI dimension brings new challenges, such as integrating data from across the enterprise and maintaining high-quality standards for GenAI initiatives. Ultimately, effective anomaly detection requires clear ownership and collaboration across the organization, progressive disclosure principles, and human expertise in data governance and critical thinking.