October 2025 Summaries

Firebase Phone Number Verification (PNV) is a developer-preview service that enables apps to securely obtain and verify a device SIM’s phone number directly through participating carriers, allowing users to sign in with a single tap rather than entering a number and SMS one-time passcode. The service works over cellular or Wi-Fi when an enabled SIM and internet connection are available, aiming to reduce onboarding friction, SMS delivery delays, phishing exposure, and artificially inflated SMS traffic costs. Because availability depends on carrier compatibility, developers can use the Firebase PNV SDK to detect support in the background and present SMS verification as a fallback when necessary. The preview launches with Deutsche Telekom in Germany and Telkomsel in Indonesia, with Firebase planning to add carriers and regions over time toward coverage comparable to SMS verification.

Firebase AI Logic offers hybrid inference for web applications, enabling AI features to use Gemini Nano on Chrome desktop when available and cloud-hosted models when needed. The preview capability supports modes that prioritize either on-device or cloud processing, allowing developers to balance offline availability, speed, cost, and model capability. Example applications include private text summarization that prefers local processing, offline image captioning using multimodal prompts, and real-time translation that routes short phrases to an on-device model while sending longer or more complex text to a cloud model such as Gemini 2.5 Flash. Firebase presents hybrid inference as a way to build resilient AI experiences for users regardless of connectivity and encourages developers to explore its documentation and participate in the Google Chrome Built-in AI Challenge 2025.

Firebase’s Model Context Protocol server, which enables AI coding agents to understand and manage Firebase projects through natural-language requests, has graduated from its experimental phase after additions such as Gemini-powered tools, broader support for products including Cloud Functions and Crashlytics, streamlined security-rule validation, and deployment prompts. Firebase also introduced an official extension for the Gemini CLI, now transitioned to Antigravity CLI, combining the MCP server with contextual instructions to support project setup, database and authentication configuration, Gemini API integration, deployment, and documentation-based assistance through commands such as `/firebase:init`, `/firebase:deploy`, and `/firebase:consult`. The extension is intended to reduce friction in terminal-based Firebase development by helping agents manage multi-step, approval-dependent tasks such as creating resources, configuring services, deploying applications, and cleanup, while Firebase invites users to provide feedback through its GitHub repository.

Firebase announced an extension originally for Gemini CLI, now transitioned to Antigravity CLI, that integrates Firebase project tools and the Firebase MCP server directly into the terminal to simplify AI feature development. Its `/firebase:init` command can configure Firebase AI Logic, register platform-specific apps, enable the Gemini Developer API, and add required SDK initialization code for web, Flutter, and Android projects. Developers can then describe a desired feature in natural language, allowing the CLI to use codebase context and provided AI Logic snippets to implement capabilities such as multimodal analysis or image generation. The announcement advises developers to expect some output variation, provide error details to the CLI, commit code frequently, and test before production deployment, while inviting feedback through the project’s GitHub repository and directing users to Firebase AI Logic documentation.

Firebase Data Connect, now renamed SQL Connect, introduces features intended to speed application development, improve console usability, and strengthen oversight of Google personnel access to customer data. Temporary databases allow developers to create schemas, add seed data, and run queries immediately while a permanent Cloud SQL instance provisions, with temporary data automatically transferred once provisioning completes. Developers can seed data manually through mutations and queries, with Gemini in the Firebase console or MCP servers available to assist, while console updates enable table-formatted query results and direct schema editing during onboarding. SQL Connect also adds Access Transparency logs, Access Approval controls, and Assured Workloads personnel restrictions, providing visibility into access events and allowing organizations to approve, deny, or limit Google personnel access based on requirements such as employee location.

Friendly Eats adds an AI-assisted feature that suggests a one-to-five-star restaurant rating while users write reviews, while preserving the reviewer’s ability to override the recommendation. The implementation uses Firebase AI Logic, which protects Gemini credentials through a proxy, supports Firebase App Check to reduce abuse, and can prioritize an on-device model such as Gemini Nano before falling back to the lower-cost Gemini 2.5 Flash Lite model. To improve reliability and simplify client handling, the app requests structured JSON output containing only the rating and uses prompt examples to establish how positive and negative review language maps to scores. Input is debounced to avoid inference requests on every keystroke, and Firestore security rules validate that submitted reviews contain text and ratings remain within the allowed one-to-five range. The accompanying TypeScript and React examples show Firebase initialization, model configuration, rating generation, UI updates, loading feedback, and manual rating controls, illustrating an approach that balances usability, cost, security, and user control; as of August 2026, Vertex AI is called Agent Platform.