November 2025 Summaries

The Pydantic AI Gateway (PAIG) is introduced as a solution to the complexities of managing large language models (LLMs), addressing issues such as API key management, rate limits, and unexpected costs, by offering a streamlined approach through a single key without the need for translation between different providers' formats. This new gateway aims to provide a superior developer experience by integrating with the existing Pydantic stack, facilitating immediate use of new features from providers like OpenAI, Anthropic, and Google without waiting for updates. PAIG emphasizes observability and control, offering features such as built-in cost limits, failover capabilities, and detailed logging for full traceability of LLM calls. The core gateway is open-source and designed to be low-latency by leveraging Cloudflare's edge compute network, while additional enterprise features like a hosted UI and SSO remain closed-source. Currently in open beta until December, PAIG is free to use as it gathers user feedback, with future pricing details to be announced closer to general availability.

Pydantic AI Gateway (PAIG) is designed to streamline the complexities of using large language models by offering a more efficient, developer-friendly alternative to traditional AI gateways. Unlike conventional gateways that employ a universal schema, PAIG allows requests to pass directly in the provider's native format, facilitating immediate access to new features from providers like OpenAI, Anthropic, and Google without waiting for updates. It offers a single key for multiple models, automated cost limits, and robust observability via integration with Pydantic Logfire, enhancing the user experience by providing detailed logs and enabling seamless transitions between models. PAIG's core is open-source, emphasizes minimal latency by using Cloudflare's global edge network, and supports enterprise features like SSO and granular permissions, making it both reliable and cost-effective for users seeking efficient LLM governance.

Pydantic AI has introduced native support for Vercel AI frontends through the VercelAIAdapter class, streamlining the integration process for building AI chatbots by eliminating the need for custom translation code between Pydantic AI and Vercel AI event formats. This integration allows developers using Starlette-based web frameworks like FastAPI to utilize the dispatch_request() method, which parses request bodies, runs agents with streaming, and encodes responses as server-sent events. For non-Starlette frameworks such as Django and Flask, developers can use individual methods of the VercelAIAdapter to maintain control over input and output. The Vercel AI Data Stream Protocol integration enables seamless event streaming between backend and frontend, transforming Pydantic AI events into Vercel AI equivalents, with minimal performance overhead. Additionally, features like the on_complete callback provide opportunities for logging, analytics, and other post-processing tasks, while Pydantic Logfire ensures cross-language observability and performance monitoring across the application stack.