Steren

stations.albertguillaumes.cat shows 3d models of many european metro stations. Here's Chatelet Les Halles in Paris:

Glad to have participated in person to the 2025 edition of Cloud Run day India! The team had prepared great deep dive sessions, quizzes, and workshops. Nice to connect with the local developer community.

Add /deploy to Gemini CLI with the Cloud Run Gemini CLI extension: $ gemini extensions install github.com/GoogleCloudP... Read more: cloud.google.com/blog/product...

Deploy your Gradio app to Cloud Run from the gradio CLI: $ gradio deploy --provider gcloud

Gemini 2.5 Flash Image (a.k.a. nano banana) is a massive leap forward, an a 171 elo point jump is huge. See nice landing page from Google DeepMind deepmind.google/models/gemin...

Node.js can now execute TypeScript nodejs.org/en/blog/rele...

A group is building a "spatial browser", i.e. a user agent that presents web content directly inside a 3D coordinate space. I love the idea. m-creativelab.github.io/jsar-runtime...

I'll be speaking at the AI For Developers meetup in San Francisco next Wednesday: - 6:30pm: Google Cloud Run for AI (Inference, agents, and MCP) - 7pm: A Python-First Approach to AI Infra, from Sandboxes to Inference - 7:30pm: Zylo-docs: The Code-first, AI-powered API Spec Tool lu.ma/qemn7akv

- Vibe coded apps go to Cloud Run - AI Agents go to Cloud Run - MCP servers go to Cloud Run Cloud Run's flexibility and on-demand nature make it a runtime of choice for these new workloads.

Thanks to Gemini CLI: - I don't write tests (manually) anymore - test coverage of my projects go up.

The Cloud Run MCP server is now listed in Docker's MCP hub. The catalog page does a great job listing the available tools and giving clear setup instructions. hub.docker.com/mcp/server/c...

Gemini CLI and I added an anaglyph mode to my <stereo-img> web component, for non-VR users. stereo-img.steren.fr

Previously, we waited for a language runtime to be Stable or LTS before launching it on Cloud Run. We are changing that: Go 1.25 (today "Release Candidate") and Node.js 24 ("Current") are available in Preview today. They will GA once Stable and "Active LTS". $ gcloud run deploy

Right now: 10,000 homes sending 60MW to the grid as one Virtual Power Plant

Carbon numbers for Google Cloud regions, 2024 edition. - Details: cloud.google.com/sustainabili... - CSV: github.com/GoogleCloudP...

I'd like to thank TikTok and Gen Z for a potential new Radiohead tour (and one can hope) album.

The Gemini CLI roadmap is now public github.com/orgs/google-... 👀 2 Cloud Run-related items: Start background agent either locally, by deploying to Cloud Run, or through Jules integrations: github.com/google-gemin... Deploy to Cloud Run with /deploy github.com/google-gemin...

In an alternate timeline, we would have flown from San Francisco to Paris in 4.5 hours today.

Announcing Docker, Inc compose support for Google Cloud Run: - Define your app in "compose.yaml" - Run locally with "docker compose up" - Deploy with "gcloud run compose up" Docker blog: www.docker.com/blog/build-a... GCP blog: cloud.google.com/blog/product... Video: youtu.be/KHM3YGkLOxo?...

Gemini API is getting a Batch Mode, to run within a 24-hour window. Priced 50% lower and with higher limits. developers.googleblog.com/en/scale-you...

GitHub PR review tool is way behind Google's internal code review tool. For performance, usability, and features

cooking something

Gemini CLI can also help you pass time when your containers are building

Nice to see that ElectricityMaps has a "Carbon aware scheduler" API endpoint. Given a job duration, start and end window, and set of locations (which can be GCP regions), the API returns the optimal execution time and location. portal.electricitymaps.com/docs/api#car...

Serverless Gemma 3n with Ollama and Cloud Run in two lines: $ gcloud run deploy --image ollama/ollama --port 11434 --gpu 1 $ OLLAMA_HOST=[...] ollama run gemma3n

Yesterday, we introduced Cloud Run worker pools, designed for continuous background processing , ideal for pull-based workloads. Today, we are introducing an open source autoscaler for Kafka consumers. Read the blog post: cloud.google.com/blog/product...

npm i -g @google/gemini-cli

The Cloud Run and Firebase MCP servers are now listed in the list of official MCP servers, alongside other very useful ones like GitHub. github.com/modelcontext...

Today at work

Build and Deploy a remote MCP server to Google Cloud Run in under 10 Minutes: cloud.google.com/blog/topics/...

Gemini 2.5 Pro: GA (input: $1.25, output: $10) Gemini 2.5 Flash: GA (input: $0.3, output: $2.5) Gemini 2.5 Flash-Lite: Preview (input: $0.1, output: $0.4) All support text, image, video, audio, Google Search, code execution and have a 1M context.

Gemini got it right

$25k for the Project Starline device (now named Google Beam) Still one of the most impressive technology and product I've ever experienced. You are present, pay attention, pick up subtle facial clues... and just forget you are in a remote call. Businesses should not hesitate.

Go home LinkedIn. You are drunk. -> Muted

Looks like macOS is getting a built-in way to run containers with a `container` command.

And that's a wrap for "Sons of The Forest". Still no can opener though.

My new benchmark for vibe coding tools: "A The Witness panel generator" So far, no tool one shots it. Tested: Gemini, AI Studio, Firebase Studio, v0, Bolt, Mocha Expectations: - basic style (circle and line on a grid) - mouse interactivity - puzzles with actual rules

Gemini Diffusion is a step function in terms of speed. Example: vibe coding a portfolio in 3s. Traditional LLMs generate text sequentially. Diffusion models start with random 'noise' and iteratively refine it (It's how image models work, now applied to text). deepmind.google.com/frontiers/ge...

The Cloud Run MCP server now allows AI agents to retrieve logs of a Cloud Run service, useful for debugging. Thanks Christoph for the PR! github.com/GoogleCloudP...

Engineers should code Managers should code Designers should code PMs should code Directors should code VPs should code CEOs should code

Lindt tier list.

"from ~ 20 minutes to 20 seconds" is the kind of improvement I love reading in the Cloud Run's team chatroom. Many such cases in the past (e.g. GPUs). Can't wait to share what this latest one is about.

The future arrived at my doorstep.

We've published documentation for how to host MCP servers on Cloud Run: cloud.google.com/run/docs/hos... It covers how to host remote streamable HTTP MCP servers. And basic guidance for how to authenticate local MCP clients as well as MCP clients running on Cloud Run.

Day 25 surviving Sons of The Forest: We explored most of the island. Our backpack is loaded with all sorts of tools, weapons, and materials. We have gold plated our katana and found an ancient armor deep in a cave full of daemons. Only one item is missing: a can opener.

For this Node.js project, I decided to use npm modules instead of writing everything from scratch like I usually do. I regret my decision: my program takes 20s to start... because of module loading. Am I supposed to use a bundler? How do you profile modules load times to identify the culprits?

TIL, Home Assistant has been donated to Open Home Foundation, a non-profit in Switzerland www.openhomefoundation.org/blog/announc...