Hello Datafan!

I spent nine early mornings animating this video about one of the most recent dashboard apps from Streamlit Co-founder Thiago Texeira.

It's an app packed with layout, caching and callback tips, that we tend to forget easily. So I made a video about it!

​

But I’ve found something that remembers those best practices for me ⬇️

✨ The best Streamlit coding agents

Tired of your agent forgetting to cache data or ignoring session_state?

What if you could disclose Streamlit best performance practices as context to Claude Code / Gemini CLI / Codex when it needs it the most?

The Streamlit dev team just released a performance agent skill to explain coding agents how to optimize Streamlit apps, straight out of the dev/product team. Add it to your .claude or .gemini or .agents /skills folder so they can load it on demand.

In fact, there is an entire repo of official agent skills to help your agent. There are skills that range from implenting Streamlit chatbots to building Custom Components in the version 2 template.

🔥 So good they rewrite my own components

To put it to the test, I pulled the custom component v2 skill, and cloned the template and streamlit-bokeh, since they are the best documented Components Resources, next to my streamlit-echarts component.

Then I asked Antigravity to cross-analyze all those Streamlit official best practices to upgrade streamlit-echarts to the version 2 template and ECharts v6, all while removing the React dependency from the library.

And nooowww, you're the very first to hear about this...I'm a few lines of Claude/Gemini adversarial code away from releasing the newest version of streamlit-echarts 0.6.0.

Comes with ECharts v6 features, React out of the way, more animations enabled, and more selection options!

Moving forward, I won’t be the main contributor to the codebase—coding agents will handle the grunt work. I’ll focus on the prompting, context-harnessing, and checking for Typescript hallucinations, I'm still in the driving seat.

This lets me get back to making videos about ECharts options which I've been wanting to do for a while!

😱 Fullstack agentic coding?

Just in! FastAPI also has its official agent skills!

Since Streamlit has experimental support for Starlette and FastAPI - you did see my latest video about Starlette right?

You are now able to fully benefit from agent skills on both the Streamlit frontend and FastAPI backend, enabling you to quickly build fullstack demos.

I feel that's a video idea in itself 📝

✒️ My current work life

My work days are full of Gemini CLI workflows to quickly build product demos, for once without Streamlit 😅 here are some for your inspiration, I should write a blogpost about each:

• Agentic search in product docs: A project comprising of command-line tools to download context from Confluence docs, JIRA boards, Office documents (thank you markitdown for converting those to Markdown files) and Gitlab source code. Agents use those CLI tools on demand to tell me which documented product features to bring to which customer demo
• GPU Orchestration: Managing a Kyutai TTS demo on a Google Compute Engine VM, using Gemini CLI to generate all my gcloud, docker, and CUDA commands.
• Salesforce Demo Interactive Writing Guide: I know nothing about Salesforce. Luckily Gemini is good at Salesforce, and when it's not good I send it off to do Web Searches to find the info.
  I have a Gemini CLI chat -> material mkdocs Markdown files + a few Python file management scripts -> Documentation Guide of Salesforce Demo -> Me trying out the guide on a Salesforce Dev platform -> Ranting to Gemini CLI that the Guide didn't use the correct button so it fixes the guide...

🔥 What's the next video about?

Haven't decided yet because of the lack of sleep. Those early mornings do take a hit on me ahah

Depends between the streamlit-echarts announcement, Streamlit Agent Skills, or an AMA because I reached 1M views on my Youtube channel ?! Let me know :)

Have a nice day!