The paper announces a suite of new capabilities for Firebase, transforming it into an end-to-end platform designed to accelerate the complete application lifecycle, particularly focusing on AI application development. This initiative aims to address the increasing complexity faced by developers due to the rapid evolution of full-stack development and the rise of generative AI.

The core offering is Firebase Studio, introduced in preview as a cloud-based, agentic development environment powered by Gemini. Firebase Studio integrates existing Firebase services with AI capabilities, providing a unified space for creating and publishing production-quality AI applications. For new applications, developers can utilize over 60 pre-built templates or leverage the App Prototyping agent. This agent assists with designing application components such as the User Interface (UI), Application Programming Interface (API) schema, and AI flows, all through natural language prompts, images, drawing tools, and screenshots. Prototypes can be iteratively refined through continued prompting and then directly deployed to Firebase App Hosting for feedback collection or experimentation, with usage monitoring available at a glance or in the Firebase Console.

From the prototyping stage, applications can be seamlessly opened within a Firebase Studio coding workspace. These workspaces allow for refinement of architecture and expansion of features for production deployment. Key capabilities within the coding workspaces include:

• Simplified coding workflows: Gemini in Firebase provides assistance for various coding tasks, including debugging, testing, refactoring, code explanation, and documentation.
• Enhancing existing applications: Developers can import existing codebases from local machines or git-based repositories (GitHub, GitLab, Bitbucket) and create custom templates for preferred technology stacks.
• Creating full-stack experiences: The environment supports customization and evolution of all application aspects, from AI model inference, agents, and Retrieval Augmented Generation (RAG) to the user experience, business logic, databases, and integration of external tools like APIs and microservices.
• Familiar tool integration: Workspaces support custom configurations, system tools, extensions, and environment variables, with access to thousands of extensions from the Open VSX Registry.
• Flexible deployment options: Applications can be configured to run on Google Cloud with built-in integrations to Firebase backend services and Google Cloud Run, or deployed on custom infrastructure.

Firebase Studio offers free workspaces during its preview, with additional capacity for Google Developer Program members.

Beyond Firebase Studio, the paper highlights advanced agentic developer tools:

• Gemini Code Assist agents: Accessible early access within Firebase Studio, these include a Migration agent (e.g., for Java code version migration), an AI Testing agent (for adversarial tests to uncover and fix potentially harmful AI model outputs), and a Code Documentation agent (providing a wiki-style knowledge base for code onboarding).
• App Testing agent in Firebase App Distribution: This agent, now in preview for Android, simulates real-world user interactions. Developers can define a goal (e.g., "Find a trip to Greece"), and the agent, powered by Gemini, formulates a plan, executes it on virtual or physical devices by navigating the UI, and provides detailed pass/fail results with rationales and visual traces of its path.

For building new AI application experiences, Firebase introduces:

• Expanded language support for Genkit: This framework, designed to simplify the development, testing, and monitoring of AI features, now offers early support for Python and expanded support for Go. Genkit facilitates building agentic experiences with features like structured output, tool calling, human-in-the-loop interactions, RAG, Model Context Protocol (MCP), and multi-model orchestration. It integrates with various models, including Gemini models, Imagen 3, Vertex Model Garden (Llama, Mistral), self-hosted models via Ollama, and third-party models via community plugins.
• Vertex AI in Firebase: This service allows developers to integrate generative AI into applications via a streamlined, secure SDK. Building upon existing support for Gemini and Imagen 3 models (Imagen 3 and Imagen 3 Fast) for image generation across Android, iOS, Flutter, and Web, new support for the Live API for Gemini models is introduced. This enables more conversational interactions, such as allowing applications to process audio questions and provide real-time responses.

To accelerate modern, data-driven applications, two key components are now Generally Available (GA):

• Firebase Data Connect: This offers robust reliability through Google Cloud SQL for PostgreSQL, coupled with instant GraphQL APIs and type-safe SDKs. It supports complex applications like social media platforms, e-commerce, and personalized recommendations with built-in vector search. Enhancements include Gemini-powered automatic generation of Data Connect schemas, queries, mutations, and client SDKs; expanded query capabilities with native aggregation, atomic data modifications, and transactions with server value expressions; and tight integration with web frameworks via generated type-safe hooks and components.
• Firebase App Hosting: An opinionated, git-centric hosting solution for modern full-stack web applications, it manages the entire application stack from build to CDN and server-side rendering, triggered by GitHub pushes. Built on enterprise-grade Google Cloud services (Cloud Build, Cloud Run, Cloud CDN), new features include a local emulator and improved error messages for build troubleshooting, a new monitoring dashboard for performance and health insights with instant rollback capabilities, and the ability to connect to a Virtual Private Cloud (VPC) for accessing non-public backend services (e.g., Cloud Memorystore, non-Firebase databases).