iiinova — where your digital self keeps evolving

Guillaume Laforge @glaforge.dev · 29/09/2026

Maybe "for hours again", so that it rhymes better?

210

Guillaume Laforge @glaforge.dev · 28/09/2026

Instead of one monolithic prompt (30-40s), LangChain4j runs 5 parallel day agents on Java 25 virtual threads via `ParallelMapperAgent`. Run time dropped to ~5s, costing ~1¢. A guardrail agent catches injections early, and deterministic Java code eliminates room collisions.

210

Guillaume Laforge @glaforge.dev · 28/09/2026

Planning your #Devoxx Belgium schedule? With 200+ talks across 8 parallel tracks, picking sessions takes time. I built a personalized agenda generator using #Micronaut, #LangChain4j, and #Gemini 3.5 Flash-Lite on #Java 25 virtual threads! Here is how it works 🧵👇

152

Guillaume Laforge @glaforge.dev · 25/09/2026

At least, it should say that it's not a cat!

010

Guillaume Laforge @glaforge.dev · 25/09/2026

Once you've picked up a picture or uploaded your own, the #DiffusionGemma model running on #NVidia GPUs will tell you if the photo contains a cat, what kind of fur pattern it has, and how fluffy it is:

210

Guillaume Laforge @glaforge.dev · 25/09/2026

Is it a 😺 or not? System 1 models like @typesafeai #jev classify, evaluate truthiness, or give a score. I built an app in #java w/ @micronautfw & @langchain4j hosted on #CloudRun to discover if a picture depicts a cat or not, thanks to #DiffusionGemma purrfect-match.cloud.run/

231

Reposted by Guillaume Laforge

Astropierre @astropierre.com · 24/09/2026

Le saviez-vous ? 99,95% de votre masse est contenue dans le noyau des atomes qui composent votre corps. Mais 99% de la masse de ces noyaux est... de l'énergie (les quarks ne comptent que pour 1%). Conclusion, la masse totale de votre corps est à 98,95%... de l'énergie (et 1,05% de particules).

static.klipy.com

Amazing Joey King

ALT: Amazing Joey King

5032170

Bringing it to #Java with @LangChain4j.dev: LangChain4j was adding StructuredDecisionModel for text. I proposed & implemented native multimodal support. In Java, pass ImageContent with typed Question models (Choice, Noul, Score) for clean, zero-shot visual classification!

What about images? DiffusionGemma includes a SigLIP vision tower. I turned vision back on in Cloud Run and sent a photo of my cat: 😺 Is it a cat? (99.99%) 👁️ Eye color: Blue (99.99%) ☁️ Fluffiness: Extra fluffy All three questions evaluated concurrently in just 129 ms! 🕑

100

taeold has 80 repositories available. Follow their code on GitHub.

The open-source foundation: Matt Mastracci (github.com/mmastrac) built djev on #DeepMind's DiffusionGemma. Using Daniel Lee's recipe (github.com/taeold), I deployed it on Cloud Run with NVIDIA Blackwell GPUs. With min-instances=0, it scales to zero idle (answering in ~140 ms!)

github.com

taeold - Overview

100

Instead of generating text token by token, "System 1" decision models (like Jev) act as fast, deterministic decision engines. In a single forward pass, they output calibrated probabilities. Here's how I turned DiffusionGemma into a multimodal visual classifier in Java! 🧵👇

1102

Designing a custom voice in Java takes just a few lines: Voice voice = client.createVoice(Voice.prompted( "A calm, deep French voice, speaking slowly", "French Philosopher", "Gemini 3.8 Flash TTS" )); Pass voice.id() to VoiceConfig. Done.

I just released v2.0.2 of my Gemini Interactions Java SDK to support these new TTS capabilities: • Voice.prompted(...) for prompt-based voice design • Voice.replicated(...) for voice cloning • Content.speech(text, style) for emotional steering • Direct audio preview access

100

Create your own voices with Gemini 3.8 text-to-speech

To hear what this sounds like in practice, watch this demo video: 👉 www.youtube.com/watch?v=FL6... It gives a great taste of how natural the voice quality is, and how prompt-based voice design lets you shape distinct vocal personas with realistic cadence.

youtube.com

Design entirely new vocal personas from scratch for gaming, immersi...

100

Text-to-speech has reached a whole new level of quality and personalization. With Google's Gemini 3.8 Flash TTS, you can design custom voices from text prompts, steer delivery nuances (whispers, pacing, emotion), and clone voices. And you can do it all in #Java! 🧵👇

152

Antigravity Java SDK v0.2.17 also brings: • Subagents with dedicated model targeting (route subtasks to smaller models) • Task scheduling tools (schedule & manage_task) • Tool argument rewriting via hooks • Synchronized native harness across macOS, Linux & Windows

Step 2: Connect from Java Configure your agent with LiteRTAgentConfig pointing to your local model path and port 9379. The SDK manages the harness lifecycle and connects to LiteRT-LM's local endpoint. Call agent.chat() as usual, but completely on-device!

100

Step 1: LiteRT-LM setup Install the on-device runtime: pip install litert-lm Download a Gemma model: litert-lm import --model-id litert-community/gemma-4-E2B-it-litert-lm --output-dir ~/.litert-lm/models/gemma4-e2b/ Serve locally: litert-lm serve --port 9379

100

Why run local models for agents? • Zero API costs: run repetitive tasks and test assertions without racking up tokens • Data privacy: proprietary code and docs stay on your machine • Offline capable: build and test without internet • Low latency: no roundtrips over the wire

100

Google Antigravity (@antigravity) on X

The #antigravity team announced running local Gemma models on-device with LiteRT. Good news for #Java developers: you can do this in Java too! Antigravity Java SDK v0.2.17 is out, reaching full parity with Python SDK v0.1.18. Walkthrough 🧵👇 glaforge.dev/posts/2026/... x.com/antigravity...

x.com

Run open models like Gemma 4 completely offline in the Antigravity SDK. Built on Google AI Edge’s LiteRT, you can now run Gemma 4 directly on your local GPU. Zero API costs, total data privacy, a…

182

Reposted by Guillaume Laforge

Olivier @olivierleplus.bsky.social · 23/09/2026

Not every cool new feature in tech is about AI 😉 I wrote a small article about the "new" CSS feature that FINALLY allows us to customize <select> elements without using JavaScript \o/ builder.aws.com/content/3Jgb...

builder.aws.com

Customizable Select: The <select> dropdown FINALLY catches up with CSS | AWS Builder Center

Style native dropdowns with CSS, from the picker and checkmark to options with rich HTML. Explore current browser support and three practical examples that keep the browser’s built-in selection behavi...

062

Simon Willison @simonwillison.net · 23/09/2026

Indeed. No hallucination, respecting the shape. But the value inside can indeed be wrong.

000

Reposted by Guillaume Laforge

The new Gemini 3.8 TTS models are super-cheap and can generate conversations between multiple voices (from 2,000+, or you can clone your own) - I built a little playground UI for it, then had Claude knock up a script where two pelicans debate moving to Pacifica Pier simonwillison.net/2026/Sep/23/...

9847

GitHub - glaforge/antigravity-java-sdk: An unofficial Java port of the Antigravity AI Agent SDK.

All 8 samples: 👉 github.com/glaforge/ag... SDK repo: 👉 github.com/glaforge/an... What use cases would you build with Antigravity (Java or Python)? PR review bots in GitHub Actions, CI audits, internal tools? Let me know!

github.com

An unofficial Java port of the Antigravity AI Agent SDK. - glaforge/antigravity-java-sdk

000

5️⃣ GitHub PR Review Agent (Main Demo) Combines custom GitHub tools, review rubrics via Agent Skills, subagents, and lifecycle hooks to compare two competing PRs in LangChain4j, stream thoughts live, and produce an objective review report.

100

Agent Skills Overview - Agent Skills

4️⃣ Open Agent Skills & Session State Load folder-based skills (SKILL.md) for progressive disclosure: keep prompts lean until specialized knowledge is needed. Plus, inject ToolContext into tool methods to store session-scoped state without polluting LLM schemas.

agentskills.io

A standardized way to give AI agents new capabilities and expertise.

120

3️⃣ Deny-by-default Security Policies Autonomous agents shouldn't have unrestricted privileges. Configure guardrails with Policies.denyIf(...) to intercept and block destructive operations before execution, allow safe read tools, and fall back to denyAll().

100

2️⃣ Real-time Streaming & Thinking Modern Gemini models think before answering. With chatStream(), you stream tokens as they arrive while separating the model's internal chain-of-thought (thoughtsDelta) from the final response text (textDelta). Great for transparent UIs.

100

1️⃣ Local Java Tools & Structured Output Annotate Java methods with @Tool and @Param to let the agent call your code. With .finishToolSchema(Record.class), the SDK derives the JSON Schema and deserializes the response directly into your Java record. Zero manual parsing.

200

Antigravity SDK for the Java Developer

Antigravity powers agents across the CLI, IDE, and desktop manager. The Java SDK wraps the native Go execution harness with a fluent, type-safe API for #Gemini models. Browse the 20-slide deck here: speakerdeck.com/glaforge/an...

speakerdeck.com

Antigravity is a family of agentic tools, comprising Antigravity 2.0 (the agent manager desktop application), Antigravity CLI (to work in the terminal),…

Gave a talk at #GDG Cloud Paris about the #antigravity SDK for #Java! While the official SDK is in Python, I built an unofficial Java SDK to bring agent automations to the JVM. Article & 8 code samples 🧵👇 glaforge.dev/talks/2026/...

121

paulk-asert.bsky.social @paulk-asert.bsky.social · 18/09/2026

Looking forward to talk about loop engineering with @wietsevenema.eu at #Devoxx!

020

Reposted by Guillaume Laforge

Version 6.0.0-RC-3 of @groovy.apache.org is available for download with 11 fixes and improvements! Thanks to all involved! Downloads: groovy.apache.org/download.html Groovy 6 Release notes: groovy-lang.org/releasenotes... #groovylang @apache.org

022

Guillaume Laforge @glaforge.dev · 18/09/2026

Pour ceux qui suivent l'actualité des #LLMs (avec la sortie de #jev) et qui aiment #DannyBoon... 😆

100

Guillaume Laforge @glaforge.dev · 16/09/2026

Ah ouais tu ressembles trop à Steve Jobs comme ça !!! J'adore !!!

140

Guillaume Laforge @glaforge.dev · 16/09/2026

I'm sure @sdkman.io will catch up. Did you open a request?