Genkit

On-device AI brings zero token costs, offline support, and unbeatable privacy. The cloud brings high-reasoning ceilings and unlimited world knowledge. Real hybrid AI is about routing requests dynamically between them. Google Developer Expert Sasaha Denisov just published a must-read architectural guide on building hybrid AI in Flutter using Genkit Dart and genkit_flutter_gemma. 💡 Level 1: Simple ternary swapping driven by app context (e.g., keeping PII local). 🧠 Level 2: Using a fast, local 1B classifier model to decide query complexity on-device. 🛡️ Level 3: Leveraging Genkit middleware to gracefully handle transient failures. 🔄 Level 4: Building a custom fallback service to degrade gracefully from cloud to local. With Genkit's provider-neutral design, switching between cloud Gemini and a local Gemma model is as simple as changing a single `model:` reference in your code. https://lnkd.in/gdw7BR-U

Genkit JS & CLI v1.36.0 is officially live! 🚀 We’ve just released a fresh set of updates to enhance your developer experience, bring new model capabilities, and harden your production AI workflows. Here is what is new in this release: 💻 New CLI Commands: Retrieve trace data directly from your terminal with the new `trace:get` and `trace:list` commands, perfect for debugging workflows without leaving the command line. 🧠 Antigravity Agent Model: Added support for the Antigravity agent model in the Google GenAI plugin. ☁️ Vertex AI Updates: Added support for `-latest` model versions to ensure you always build with the most up-to-date models. 🤖 Anthropic SDK Upgrades: The Anthropic SDK is now updated to ^0.97.1 to gracefully handle `input_json_delta` stream events instead of throwing errors. We’ve also shipped critical bug fixes, including resolved Claude thinking support in Vertex AI Model Garden and more robust tool response handling in the filesystem middleware. Check out the full release notes and upgrade your projects today: https://lnkd.in/gRFHqb_v

Dart developers can now build full-stack AI applications with official Genkit Dart! 🎯 But moving AI from prototype to production means thinking about security and abuse prevention. In this video, Nohe demonstrates how to secure a Genkit Dart backend (using a Shelf server) for a virtual try-on application. 🔒 Device Attestation: Using Firebase App Check to verify legitimate clients. 🔐 User Authorization: Checking custom claims in authorization headers. ⏱️ Rate Limiting: Using Firestore aggregation queries to keep costs in check. 🛡️ Prompt Injection Prevention: Restricting inputs to product IDs rather than arbitrary URLs. Watch the full video to learn how to harden your Dart AI apps: https://lnkd.in/gpF_9gRf

🚀 Genkit JS & CLI v1.35.0 is officially live! We are continuing to expand model support and harden production reliability. Here is what is new in this release: ✨ Gemini 3.5 Support: We have added support for Gemini 3.5 and Gemini 3.5 Flash in the Google GenAI plugin. 🔍 Better Debug Traces: Improved trace visibility to help you debug complex AI workflows. 🛡️ Smarter Error Handling Check out the full release notes and upgrade your projects today: https://lnkd.in/gpBG7FcU

Building production-ready AI agents requires more than just powerful models and careful prompting. You need reliability, safety, and fine-grained control. Genkit is our open-source framework for building full-stack, agentic applications. With the new middleware system, you can now use composable hooks to intercept generation calls and inject custom behaviors. Here is what is available today in TypeScript, Go, and Dart: 🔄 Automatic Retries: Handle transient errors with exponential backoff. 🛡️ Tool Approval: Implement human-in-the-loop safety for autonomous agents. 📦 Filesystem Access: Give your models scoped, safe access to the local filesystem. You can also build and stack your own custom middleware to enforce business rules or security policies. Read the full announcement on the Google Developers Blog https://lnkd.in/gQnbJPsX

We love seeing our community of educators grow for Genkit!

Today, @intflows/genkit-guard gets a major upgrade. Version 0.0.8-alpha.1 uses the new openai/privacy-filter (q4) model for local PII detection. This model from OpenAI is designed specifically for data sanitization and secret detection. If you are building with Genkit and need to ensure PII or Secrets don't hit the cloud based LLM, give v0.0.8-alpha.1 a spin. The Specs: -> Local Inference: Zero latency/privacy risk from external PII API's. -> Secret Detection: Catches passwords and keys before they leak. -> Seamless: Just drop it into your ai.generate use array. Try it now: npm i @intflows/genkit-guard@0.0.8-alpha.1 #FirebaseGenkit #Genkit #CyberSecurity #HuggingFace #TransformersJS Hugging Face Genkit

🚀 Genkit JS v1.34.0 is here with major updates and critical fixes! Here’s what’s new in this release: ✨ PayGo support for priority and flex models. ✨ Support for new models including veo-3.1-lite-generate-001 and Gemini 3.1 Flash Lite. ✨ Support for MCP multipart tools. ✨ Support for new deep research models. ⚠️ IMPORTANT: This release handles upcoming breaking changes for the Gemini API's Interactions endpoint. If you are using deep research or Lyria3 models, you must update to this version to ensure your code works past May 26th. Full release notes: https://lnkd.in/g-xwaSGa

🚀 Genkit Dart v0.13.2 is now available! 🎯 This release focuses on improving the developer experience by adding enhanced support for Middleware in the Developer UI. Now you can more easily inspect and debug your middleware-wrapped generation loops and tool calls directly in the UI. Check out the full release notes and upgrade your projects: https://lnkd.in/gqXzycd2