Tag: Google

FunctionGemma is a specialized AI model for function calling. This post explains why fine-tuning is key to resolving tool selection ambiguity (e.g., internal vs. Google search) and achieving ultra-specialization, transforming it into a strict, enterprise-compliant agent. A case study demonstrates the improved logic. It also introduces the "FunctionGemma Tuning Lab," a no-code demo on Hugging Face Spaces, which streamlines the entire fine-tuning process for developers.

Google Cloud enables end-to-end confidential applications, protecting sensitive data 'in-use' with hardware isolation. The solution combines Confidential Space (TEE/attestation), Oak Functions (private sandbox), and Oak Session (attested end-to-end encryption for scale). This framework anchors user trust in open-source components, proving confidentiality for sensitive workloads like proprietary GenAI models, even when running behind untrusted load balancers.

The JAX AI Stack is a modular, industrial-grade, end-to-end machine learning platform built on the core JAX library, co-designed with Cloud TPUs. It features key components like JAX, Flax, Optax, and Orbax for foundational model development, plus an extended ecosystem for the full ML lifecycle and production. This integration provides a powerful, scalable foundation for AI development, delivering significant performance advantages.

Introducing Google Antigravity, a new agentic development platform for orchestrating code. It combines an AI-powered Editor View with a Manager Surface to deploy agents that autonomously plan, execute, and verify complex tasks across your editor, terminal, and browser. Agents communicate progress via Artifacts (screenshots, recordings) for easy verification. Available now in public preview.

LiteRT's new Qualcomm AI Engine Direct (QNN) Accelerator unlocks dedicated NPU power for on-device GenAI on Android. It offers a unified mobile deployment workflow, SOTA performance (up to 100x speedup over CPU), and full model delegation. This enables smooth, real-time AI experiences, with FastVLM-0.5B achieving over 11,000 tokens/sec prefill on Snapdragon 8 Elite Gen 5 NPU.

Gemini 3 is available via API with updates for developers: new `thinking_level` for depth control, `media_resolution` for multimodal processing, and enforced `Thought Signatures` for agentic workflows, especially with function calling and image generation. It also introduces combining Google Search/URL Grounding with Structured Outputs and new usage-based pricing for Grounding. Best practices, like using default temperature, are advised for optimal results.

The new Data Commons extension for the Gemini CLI makes accessing public data easier. It allows users to ask complex, natural-language questions to query Data Commons' public datasets, grounding LLM responses in authoritative sources to reduce AI hallucinations. Data Commons is an organized library of public data from sources like the UN and World Bank. The extension enables instant data analysis, exploration, and integration with other data-related extensions.

ADK introduces **Context Engineering** to scale AI agents beyond large context windows. It treats context as a compiled view over a tiered, stateful system (**Session, Memory, Artifacts**). This architecture uses explicit processors for transformation, enables efficient compaction and caching, and allows for strict, scoped context handoffs in multi-agent workflows to ensure reliability and cost-effectiveness in production.

LiteRT and MediaTek are announcing the new LiteRT NeuroPilot Accelerator. This is a ground-up successor for the TFLite NeuroPilot delegate, bringing seamless deployment experience, state-of-the-art LLM support, and advanced performance to millions of devices worldwide.

Gemini CLI's new automatic **Session Management** (v0.20.0+) saves your conversation history, tool outputs, and reasoning, providing project-specific context. Resume easily using the **Interactive Session Browser** (`/resume`) or command-line flags (`--resume`). This feature ensures you never lose your work state, capturing prompts, tool execution details, and usage stats. Customize history with cleanup policies in `settings.json`.