DEV Community: leslysandra

Andrew Ng's new open-source project, Context Hub, attempts to solve a problem every API provider has right now whether they know it or not. Coding agents are getting your API wrong.

leslysandra — Mon, 23 Mar 2026 19:41:08 +0000

Mike Chambers for AWS

Mar 10

Context Hub Has 68 APIs. Add Yours.

#ai #api #productivity

Comments 1

6 min read

Gemini 3.1 Pro & "Nano Banana 2": A Technical Review and Real-World Tests for Devs

leslysandra — Sat, 28 Feb 2026 02:51:27 +0000

While the hype around new AI models in the last week of February has everyone distracted generating "cool images" 🍌🍌🍌, the real workhorse for developers was quietly released: Gemini 3.1 Pro.

As an AI educator, I was incredibly curious about these new models, so I put them through a series of specific stress tests: advanced reasoning in ARC-AGI-2, the new 65k output token limit, and UI text consistency with Nano Banana.

If you are a dev, a data scientist, or a tech enthusiast building digital products, here is exactly what you need to know.

Gemini 3.1 Pro: Your New Logic Sidekick

Launched this February 2026, the most interesting thing about Gemini 3.1 Pro isn't just that it's "smarter." It's the architecture. Google is promising an impressive 77.1% score on ARC-AGI-2.

If you're new to AI benchmarks: ARC is the "final boss" of reasoning. Most models just memorize answers from their training data; ARC forces them to solve entirely novel patterns they have never seen before.

The Pain Point it Solves: The Output Limit

Have you ever asked an AI to refactor a 2,000-line file, only for it to cut off halfway through? It leaves you with a broken JSON, forcing you to type "continue..." while praying it doesn't lose context.

Gemini 3.1 Pro raises the output limit to a massive 65,536 tokens.

The Test: I asked it to refactor a Flask codebase (three separate Python files) in a single shot.

The prompt:
Act as a Senior Backend Architect. You are tasked with refactoring a legacy Flask application where views, models, and business logic are tightly coupled. You must migrate this entire codebase to a modern FastAPI (Asynchronous) architecture using Pydantic v2 for schema validation and SQLAlchemy 2.0 (Async) for the data layer.
Strict Requirements:
No Summarization: I need the full, complete implementation. Do not use placeholders like # ... rest of code here.
Architectural Separation: Clearly separate the code into Pydantic Models (Schemas), Database Models, and API Endpoints (Routers).
Logic Integrity: Maintain 100% of the original business logic, including complex relationships like tags, favorites, and follower/user relationships.
Modern Standards: Use Python type hints throughout, implement async/await for all DB operations, and utilize SQLAlchemy 2.0's Mapped and mapped_column syntax.
Output Format: Generate a single, massive code block using comments (e.g., # models.py, # schemas.py, # main.py) to indicate the suggested file structure for a production-ready repository.

With Gemini 3.1 Pro and the 65k tokens enabled, the generation just keeps going. By the time it finishes, it has written:

All Pydantic schemas (UserSchema, ArticleSchema…)
The asynchronous DB configuration.
All CRUD endpoints converted to async def.

(If you want to try this yourself, I used the flask-realworld-example-app repo. I fed it conduit/articles/views.py for complex route logic, conduit/articles/models.py, and conduit/user/models.py for relational database models).

Testing the Logic: The ARC-AGI-2 Challenge

To see if it actually "reasons," I used the new parameter thinking_mode='medium'. This allows the model to "think" before responding, perfectly balancing speed and intelligence.

The Test: I gave it a medium-level "trace and fill" test (essentially a hidden pattern puzzle).

The prompt:
Observe the transformation pattern between the input and output arrays. Identify the underlying rule and solve the final case.
Example 1:
Input: [1, 0, 0, 0, 2]
Output: [1, 8, 8, 8, 2]
Example 2:
Input: [0, 1, 0, 2, 0]
Output: [0, 1, 8, 2, 0]
Final Problem:
Input: [0, 0, 1, 0, 0, 0, 2]
Output: ?
Explain your reasoning.

Did you catch the hidden sequence? Are you convinced by Gemini 3.1 Pro’s result?
🤫 Psst... the hidden pattern is: Convert any zero (0) located between a 1 and a 2 into an eight (8). ⚡️⚡️⚡️

I tested this on older versions of Gemini, and the majority failed to get the correct answer. 🤓🤓🤓

Note: AI models don't 'see' these images the way we do. I passed the puzzle raw as 2D JSON matrices to prevent basic image recognition and force the model to use pure mathematical reasoning.

2. Nano Banana 2: Fast, Furious, and... Can it Read?

Let's move on to the model with the funny name 🍌🍌🍌. Nano Banana 2 (officially Gemini 3.1 Flash Image) promises Pro-level quality at Flash-level speeds.

For Frontend Devs, there are two game-changers here:

Text Rendering: Finally, an AI that doesn't write alien hieroglyphs.
Search Grounding: It can search Google before generating the image.

Try It Yourself (Hands-on Tests)

I encourage you to open Google AI Studio or the Gemini App and try this right now.

Test A: The UI Mockup (Perfect for Frontend)

Image models are notoriously bad at prototyping interfaces.
Let's test this prompt on Nano Banana 2:

Prompt: A high-fidelity mobile banking dashboard. The header says 'Total Balance: $14,250.55'. A green button at the bottom says 'Transfer Funds'. Dark mode, material design.

It generated in under 2 seconds, and the text was flawless! Imagine connecting this to your workflow, essentially "compiling" text into UI mockups in real-time for your clients.

Test B: Search Grounding (What wasn't in the dataset)

This is the killer feature. I asked for something that didn't exist when the model was trained—something based on today's breaking news (simulated).

Prompt: A realistic photo of the stage at today's tech convention, showing the leaked triangular camera design of the new VR device.

The model actively searched for the reference and generated the correct design. This opens the door to generating dynamic content based on real-time, real-world events.

Extra: One more test

To truly push Nano Banana 2's new features to the limit, I crafted a highly complex prompt. Here is the prompt and the resulting image:

Prompt: Using the two uploaded images for character reference, create a single, high-quality 4K vertical infographic titled 'The 1920s Lunar Landing.' The main subject is the person, standing on the lunar surface, dressed as a vintage 1920s-style astronaut in a polished brass helmet and detailed leather straps. They must retain their exact facial features. Next to them, integrate the cat (exactly as referenced in the second image) into its own miniature, customized vintage brass and glass bubble helmet. The cat is sitting upright near a clear, legible sign that says: 'One small step for a Flapper, one giant leap for Feline-kind.' The overall art style must be sophisticated 1920s Art Deco. The dramatic lighting from the Earth in the background must realistically reflect off both of their brass helmets, showcasing intricate texture and form.

This is a demanding prompt because it requires three distinct levels of advanced reasoning:

Parallel Subject Preservation: I uploaded two reference photos (one of me, one of my cat), and the model successfully maintained both of our exact identities.
Multitasking Lighting Logic: It had to calculate complex light reflections from the Earth onto two different brass helmets.
Complex Spatial Relationships: It logically placed the cat "next to me" and "near the sign," maintaining perfect scene coherence.

Conclusion: Where Should You Start?

Sometimes, we hesitate to adopt new tools, assuming they are "too complex" or "too expensive." The reality is that this technology is more accessible than ever.

My recommendation to start experimenting today:

If you have heavy Python scripts or complex RAG pipelines, migrate to Gemini 3.1 Pro. That 65k output token limit will give you immense peace of mind.
If you are building any app that generates images, switch to Nano Banana 2. The sheer speed combined with perfect text accuracy will finally allow you to create highly usable interfaces.

Try it yourself 😉. Here are the launch links and references to get you started:

Gemini 3.1 Pro: A smarter model for your most complex tasks https://blog.google/innovation-and-ai/models-and-research/gemini-models/gemini-3-1-pro/
Nano Banana 2: Combining Pro capabilities with lightning-fast speed https://blog.google/innovation-and-ai/technology/ai/nano-banana-2/

Last part! (3 out of 3) to Get started with Gemini CLI

leslysandra — Mon, 09 Feb 2026 19:21:49 +0000

Part 3: Mastering Gemini CLI – Content Creation, Learning, and Multimodality

leslysandra ・ Feb 9

#gemini

Part 3: Mastering Gemini CLI – Content Creation, Learning, and Multimodality

leslysandra — Mon, 09 Feb 2026 19:19:37 +0000

Welcome to the finale of our Gemini CLI series!

In Part 1, we installed the CLI and set up our environment. In Part 2, we learned about coding workflows, data analysis and extensions in workspace.

Now, we are going to have some fun. We are moving beyond simple text and code. We are going to explore multimodality (handling images, audio, and PDFs) and turn your terminal into the ultimate Personal Tutor.

If you think the command line is just for boring text, this post will change your mind.

1. Content Creation with Extensions (The "NanoBanana" Workflow)

One of Gemini's greatest strengths is that it is multimodal—it understands code, text, images, and audio natively. But how do we harness this in a terminal?

We use extensions.

Google recently introduced a robust extensions framework that lets you plug almost anything into the CLI. A popular community extension for creative content generation is NanoBanana.

This tool connects your CLI to image generation models (like gemini-2.5-flash-image), allowing you to create placeholder assets, icons, or visual concepts without leaving your code editor.

How to Connect NanoBanana

Giving your CLI "eyes for images" takes just one command.

Step 1: Install the Extension
Run this command in your terminal to pull the extension from the repository:

gemini extensions install https://github.com/gemini-cli-extensions/nanobanana

Step 2: Restart and Verify
Restart your CLI. You can now use specific slash commands like /generate or /icon.

Step 3: Generate Creative Assets
Let’s say you are building a mobile app and need a quick placeholder icon for a "Cyberpunk Todo List."

The Prompt:

"Using the NanoBanana extension, /generate an app icon for a productivity app with a cyberpunk neon aesthetic. Make it simple, vector style, on a black background."

Why this matters:
You are orchestrating creative workflows without leaving your coding environment. You become a "technical artist" straight from the command line, rapidly prototyping UI elements while you code the backend.

Ready:

2. Gemini CLI as Your Personal Tutor

The most underrated feature of Gemini 3 Flash is its massive Context Window. It can read huge files—like entire books or long PDF research papers—in seconds.

This turns the CLI into a powerful study buddy that creates active learning materials for you.

Scenario: The University Student / Self-Learner

Imagine you have a 50-page PDF called Advanced_Algorithms.pdf and you have an exam tomorrow.

Step 1: The Summary
Don't read the whole thing linearly. Ask Gemini to break it down.

gemini "Read @Advanced_Algorithms.pdf. Summarize the key concepts by chapter. Use bullet points and simple language."

Step 2: Active Recall (Flashcards)
Passive reading is inefficient. Force yourself to remember with AI-generated flashcards.

gemini "Based on @Advanced_Algorithms.pdf, generate 10 flashcards. Format them as: 'Front: [Question] | Back: [Answer]' so I can import them into Anki."

Step 3: The Mock Exam
Test your knowledge immediately.

gemini "Act as a strict professor. Create a 5-question multiple-choice quiz based on Chapter 3 of the PDF. Don't give me the answers until I try to answer them."

3. Grounding Your Knowledge with Web Search

Large Language Models (LLMs) can sometimes "hallucinate" (make things up) or rely on outdated training data. To fix this, Gemini CLI has a built-in Google Search tool (often referred to as "Grounding").

This is crucial when you are learning a new technology that came out yesterday.

Example: Learning a New Framework
If you ask standard AI about the very latest version of a library, it might give you old code.

The Prompt:

"I want to use the new features in React 19. Search the web for the official React 19 release notes and documentation. Then, explain the top 3 breaking changes and provide a code example for each."

Why this builds authority:
By ensuring the /google tool (or web search capability) is enabled in your /settings, you are proving that your code is up-to-date and fact-checked against the real world.

Conclusion: The "All-in-One" Developer

We have come a long way in this series.

Installation & Basics: We learned to navigate the CLI.
Workflow Automation: We connected to extensions, workspace and analyzed data.
Mastery: We used Extensions like NanoBanana for creativity and transformed PDFs into interactive learning materials.

The Gemini CLI isn't just a tool; it's a layer of intelligence over your entire operating system. It allows you to build faster, learn quicker, and create more—all from the comfort of your terminal.

Now, it’s your turn.
Download the CLI, install an extension, and build something amazing. Don't forget to share your creations!

Special thanks to the DeepLearning.AI course "Gemini CLI" for the inspiration for this blogpost.

@leslysandra

How to get started with Gemini CLI :)

leslysandra — Mon, 09 Feb 2026 17:04:49 +0000

How to Get Started and Build with Gemini CLI (Powered by Gemini 3 Flash)

leslysandra ・ Feb 9

#ai #cli #gemini #tutorial

Part 2: Supercharging Your Workflow with Gemini CLI

leslysandra — Mon, 09 Feb 2026 16:54:35 +0000

In Part 1, we covered how to install Gemini CLI and the basics of interacting with files using the @ symbol. Now, let’s unlock the real power of this tool.
Once you have the CLI running, you aren't limited to just chatting with text files. You can connect to databases, manage GitHub repositories, and even write directly into Google Docs.
Here are four practical ways to level up your Gemini CLI skills.

1. The Power of MCP (Model Context Protocol)

This is the most "technical" sounding part, but it is actually the most exciting.

What is MCP?
Think of the Model Context Protocol (MCP) as a "Universal USB Port" for AI.
Before MCP, if you wanted Gemini to talk to your SQL database or your Slack, you had to write complex custom code. With MCP, there is a standard way to plug these tools together.

How to Connect:
Gemini CLI acts as an MCP Client. You can run "servers" (which are just small programs that talk to your data) and Gemini can instantly interact with them.

Common MCP Servers to Try:

GitHub Server: Lets Gemini search your repos, read issues, and create pull requests.
PostgreSQL/SQLite Server: Lets Gemini run SQL queries to answer questions about your data.
Filesystem Server: Gives Gemini safe access to navigate your folders (built-in).

To get started, check the /settings command or the configuration file to add an MCP server configuration.

2. Using Extensions: The Google Docs Example

Gemini CLI allows you to install extensions to expand its capabilities beyond the terminal. A great example of this is the Google Docs integration.

Why keep your AI drafts in the terminal? You can have Gemini write them directly into a document.

How to Install Extensions:
Extensions are typically installed via npm (Node Package Manager). For example, to add Google workspace capabilities, you might look for community or official packages (always check npm search gemini-cli-extension).

Real-Life Use Case:
Imagine you have finished coding a feature and need to write documentation.

The Prompt: > "Read the code in @server.js and write a technical documentation summary. Then, create a new Google Doc named 'API Documentation' and paste the content there."
The Result: You stay in the terminal, but the work appears instantly in your Google Drive.

3. Software Development: The GitHub Workflow

This is where Gemini CLI shines for developers. By connecting Gemini to your GitHub (via MCP or extensions), you can automate the tedious parts of coding.

Implementing Features:
Instead of switching between your browser (to read the ticket) and VS Code, do it all in one place.

The Scenario: You have a bug reported in GitHub Issue #42.
The Prompt: > "Read GitHub Issue #42. Then, look at @login.py and explain why this bug is happening. Finally, suggest the code changes to fix it."

Code Reviews & PRs:
You can ask Gemini to act as a senior engineer.

The Prompt: > "I just finished writing @auth.ts. Review it for security vulnerabilities before I commit."

4. Data Analysis on the Fly

You don't need to be a Data Scientist or open Jupyter Notebooks to get quick insights. Gemini CLI can digest raw data files and give you answers.

What Data Can It Handle?

CSV Files: Excel exports, financial data.
JSON: API responses, configuration files.
Log Files: Server error logs (.txt, .log).

Example: Analyzing Sales Data
Let’s say you have a file called q1_sales.csv.

The Prompt: > "Analyze @q1_sales.csv. Calculate the total revenue for March. Also, tell me which product category had the highest growth percentage compared to February."

Why this is powerful:
Under the hood, Gemini isn't just "guessing" the numbers. It can actually write and execute a small Python script to calculate the math precisely, then show you the result. It combines the reasoning of an LLM with the precision of a calculator.

Final Thoughts

The Gemini CLI transforms your terminal from a simple command prompt into an intelligent workspace. Whether you are fixing bugs on GitHub, analyzing CSVs, or drafting docs, the goal is the same: keep your hands on the keyboard and stay in the flow.

Have you tried connecting an MCP server yet? Let me know in the comments what you’ve built!

Credits: Concepts and structure inspired by DeepLearning.AI Course: Gemini CLI.

How to Get Started and Build with Gemini CLI (Powered by Gemini 3 Flash)

leslysandra — Mon, 09 Feb 2026 16:17:09 +0000

By Lesly Zerna / @leslysandra

As developers, we are constantly looking for tools that speed up our workflow without breaking our flow state. Recently, Google made a massive splash in the open-source community by updating the Gemini CLI with their latest model, Gemini 3 Flash.

If you are tired of copy-pasting code between your browser and your IDE, or if you want to see what a true "AI Agent" looks like in your terminal, this guide is for you.

Note: Many of the concepts and examples in this post were inspired by the excellent [Insert Name of Course/Instructor Here], which I highly recommend checking out.

What is Gemini CLI?

In simple terms, Gemini CLI is a terminal application that lets you chat with Google’s Gemini AI directly from your command line.

But it is more than just a chatbot. It is an open-source AI agent powered by Gemini.

When you use ChatGPT or the Gemini web interface, you are in a "sandbox"—the AI cannot see your files or run commands unless you upload them. Gemini CLI breaks that wall. It lives inside your environment, meaning it can understand your project structure, read your code directly, and help you build faster.

Why is the "Gemini 3 Flash" update important?
Gemini 3 Flash is optimized for speed and low latency. When you are using an AI Agent that needs to perform multiple steps (like "read this file," then "analyze it," then "write a test"), speed is critical. Flash makes the CLI feel snappy and responsive.

Under the Hood: How Gemini CLI Works

You might be wondering, "How does a text box in my terminal know how to run code?"

Gemini CLI uses a concept called the ReAct Loop (Reason + Act). Here is the simple explanation of what happens when you type a command:

Thought: You ask a question (e.g., "Why is my build failing?"). The AI analyzes your request.
Tool Use: Instead of just guessing, the AI sees it has "tools" available—like reading a file or running a shell command. It decides, "I should run the build command to see the error."
Observation: The CLI runs the command, captures the output (the error message), and feeds it back to the AI.
Answer: The AI reads the error and gives you the specific fix.

This "Loop" allows the CLI to act like a real pair programmer rather than just a text generator.

Why Use the CLI? (Browser vs. Terminal)

Working from the browser is great for general questions, but for building software, the Command Line Interface (CLI) is superior. Here is why:

Direct File System Access: You don't need to copy-paste code. You can simply say, "Refactor the utils.py file," and the CLI can read it and propose changes directly.
Context Awareness: The CLI knows where it is. It understands your current working directory and the files inside it.
Automation & Scripting: You can pipe output from other commands into Gemini. For example: cat logs.txt | gemini "Find the critical error".
Built-in Tools: Gemini CLI comes battery-included with tools for:
- File System: Read, write, and list files.
- Web: Fetch URLs and search Google (grounding your code in real-time info).
- Shell: Execute terminal commands safely.
Customization & Extensions: Because it is open source, you can extend it. It supports the Model Context Protocol (MCP), allowing you to build custom tools that connect to your database or internal APIs.

Real-Life Examples and Ideas to Build

Here are a few ways you can start using Gemini CLI today to boost your productivity.

1. The "Smart" Code Reviewer

Instead of manually checking your code for style issues, use Gemini CLI to review it before you push.

Command:

gemini "Review main.py for potential bugs and clean code practices. Output the suggestions as a list."

Why it works: The CLI reads the file directly and uses Gemini’s logic to find edge cases you might have missed.

2. Instant Data Analyst

Imagine you have a messy CSV file or a log file and you need quick insights.

Scenario: You have a file called server_logs.csv.
Command:

gemini "Read server_logs.csv and tell me which IP address appears the most frequently. Then, use grep to count exactly how many times it appears."

Why it works: The agent combines its ability to write code (to analyze the CSV) with system tools (like grep) to verify the data.

3. Rapid Prototyping

Need to scaffold a new feature?

Command:

gemini "Create a new folder called 'blog-app'. Inside it, create an index.html file with a basic responsive layout and a style.css file with a dark mode theme."

Result: The CLI actually creates the folder and writes the files for you. You are ready to start coding immediately.

How to Get Started: Installation

Ready to try it? Installing Gemini CLI is straightforward. You will need Node.js installed on your machine.

(Based on the official documentation at geminicli.com)

Step 1: Open your terminal.

Step 2: Run the install command.

npm install -g @google/gemini-cli@latest

Step 3: Authenticate.
You can start the tool by simply typing:

gemini

On your first run, it will ask you to authenticate. You can usually log in directly with your Google account, which gives you a generous free tier to start experimenting.

Step 4: Enable Gemini 3 Flash.
Once inside the CLI, you want to make sure you are using the latest model.
Run this command inside the tool:

/settings

Ensure "Preview Features" is toggled to True (if required) or use /model to select the Gemini 3 Flash model for the best balance of speed and intelligence.

Conclusion

The shift from chat-in-browser to agent-in-terminal is a huge leap for developer productivity. With Gemini 3 Flash now powering the Gemini CLI, we have a tool that is fast enough to keep up with our thoughts and smart enough to handle complex tasks.

Give it a try, explore the open source code, and let me know what you build!

Special thanks to [Insert Course Name] for the inspiration on these use cases.

How to Get Started and Build with Gemini CLI (Powered by Gemini 3 Flash)

leslysandra — Sun, 01 Feb 2026 02:58:59 +0000

By Lesly Zerna / @leslysandra

As developers, we are constantly looking for tools that speed up our workflow without breaking our flow state. Last June 2025, Google made a massive splash in the open-source community by releasing the Gemini CLI and in December 2025, by updating with their latest model, Gemini 3 Flash.

If you are tired of copy-pasting code between your browser and your IDE, or if you want to see what a true "AI Agent" looks like in your terminal, this guide is for you :)

Note: Many of the concepts and examples in this post were inspired by this Gemini CLI course by DeepLearning.AI, which I highly recommend checking out, as well as checking the official Gemini CLI documentation as well as Google for devs blog.

What is Gemini CLI?

In simple terms, Gemini CLI is a terminal application that lets you chat with Google’s Gemini AI directly from your command line.

But it is more than just a chatbot. It is an open-source AI agent powered by Gemini.

Why Use the CLI? (Browser vs. Terminal)

Working from the browser is great for general questions, but for building software, the Command Line Interface (CLI) is superior. Here is why:

Direct File System Access: You don't need to copy-paste code. You can simply say, "Refactor the utils.py file," and the CLI can read it and propose changes directly.
Context Awareness: The CLI knows where it is. It understands your current working directory and the files inside it.
Automation & Scripting: You can pipe output from other commands into Gemini. For example: cat logs.txt | gemini "Find the critical error".
Built-in Tools: Gemini CLI comes battery-included with tools such as:
- File System: Read, write, and list files.
- Web: Fetch URLs and search Google (grounding your code in real-time info).
- Shell: Execute terminal commands safely.
Customization & Extensions: Because it is open source, you can extend it. It supports the Model Context Protocol (MCP), allowing you to build custom tools that connect to your database or internal APIs.

Real-Life Examples and Ideas to Build

Here are a few ways you can start using Gemini CLI today to boost your productivity.

1. The "Smart" Code Reviewer

Instead of manually checking your code for style issues, use Gemini CLI to review it before you push.

Command:

gemini "Review @main.py for potential bugs and clean code practices. Output the suggestions as a list."

Why it works: The CLI reads the file directly and uses Gemini’s logic to find edge cases you might have missed.

2. Instant Data Analyst

Imagine you have a messy CSV file or a log file and you need quick insights.

Scenario: You have a file called server_logs.csv.
Command:

gemini "Read @server_logs.csv and tell me which IP address appears the most frequently. Then, use grep to count exactly how many times it appears."

Why it works: The agent combines its ability to write code (to analyze the CSV) with system tools (like grep) to verify the data.

3. Rapid Prototyping

Need to scaffold a new feature?

Command:

gemini "Create a new folder called 'blog-app'. Inside it, create an index.html file with a basic responsive layout and a style.css file with a dark mode theme."

Result: The CLI actually creates the folder and writes the files for you. You are ready to start coding immediately.

How to Get Started: Installation

Ready to try it? Installing Gemini CLI is straightforward. You will need Node.js installed on your machine.

(More details, see the official documentation at geminicli.com)

Step 1: Open your terminal.

Step 2: Run the install command.

npm install -g @google/gemini-cli@latest

Step 3: Authenticate.
You can start the tool by simply typing:

gemini

On your first run, it will ask you to authenticate. You can usually log in directly with your Google account, which gives you a generous free tier to start experimenting.

And once it is installed you can see it running in your terminal:

Step 4: Enable the Latest Features (Crucial Step!)
To use the powerful Gemini 3 Flash model, you need to enable preview features.

Type /settings in the CLI.
Look for "Preview Features".
Set this to "True". This ensures you are running on the latest, fastest, and most capable models available.

Under the Hood: How Gemini CLI Works

You might be wondering, "How does a text box in my terminal know how to run code?"

Gemini CLI uses a concept called the ReAct Loop (Reason + Act). Here is the simple explanation of what happens when you type a command:

Thought: You ask a question (e.g., "Why is my build failing?"). The AI analyzes your request.
Tool Use: Instead of just guessing, the AI sees it has "tools" available—like reading a file or running a shell command. It decides, "I should run the build command to see the error."
Observation: The CLI runs the command, captures the output (the error message), and feeds it back to the AI.
Answer: The AI reads the error and gives you the specific fix.

This "Loop" allows the CLI to act like a real pair programmer rather than just a text generator.

Navigating the Terminal: Essential Commands

Once you have installed Gemini CLI, you aren't just typing into a void. There are specific commands to help you control the agent.

/help: The most important command. Use this to see what the agent can do and how to interact with it.
/settings: Use this to explore your setup. This is where you configure your Assistant settings and change the Mode.
/docs: Opens the documentation if you need to read up on advanced features.

How to Reference Your Files

This is the superpower of the CLI. You don't need to copy-paste code. To ask Gemini about a specific file, simply use the @ symbol.

Example:

"Can you explain the logic inside @main.py and suggest improvements?"

The agent will read that specific file and base its answer on the code inside it.

Understanding Context and `GEMINI.md`

What is Context?
Context is simply a collection of information that the AI agent uses to understand your specific request. It’s the "background knowledge" the AI needs so you don't have to repeat yourself.

The Magic of `GEMINI.md`

You can teach the CLI about your project by creating a special file called GEMINI.md. Think of this as a "ReadMe for the AI."

If you place a GEMINI.md file in your folder, the CLI automatically reads it to understand your project rules, coding style, or architecture. You can place this file in:

The Global Directory: For rules that apply to all your projects.
The Current Working Folder: For project-specific rules.
Subdirectories: For rules specific to that module.

Managing Memory

You can verify what the AI currently "knows" or add new things to its memory manually using these commands:

Check Context:
```
/memory show
```
This will list everything currently in the agent's context.

Add Context Manually:

/memory add "Remember that we are using Python 3.11 for this project"

Or you can point it to a specific context file:

/memory add path/to/another/gemini.md

Already excited about the topic? This is just the intro, we'll explore more about CLI Extensions, uses for Software Development, Data Analysis, MCP and more in the coming blogs!

Conclusion (This first part!)

The shift from chat-in-browser to agent-in-terminal is a huge leap for developer productivity. With Gemini 3 Flash now powering the Gemini CLI, and the ability to customize Context using GEMINI.md, we have a tool that is fast enough to keep up with our thoughts and smart enough to handle complex tasks.

Give it a try, explore the open source code, and let me know what you build!

More blogs coming up soon! Subscribe and stay tuned!, Lesly

How to Get Started and Build with Gemini CLI (Powered by Gemini 3 Flash)

leslysandra — Sun, 01 Feb 2026 02:58:59 +0000

By Lesly Zerna / @leslysandra

If you are tired of copy-pasting code between your browser and your IDE, or if you want to see what a true "AI Agent" looks like in your terminal, this guide is for you :)

What is Gemini CLI?

In simple terms, Gemini CLI is a terminal application that lets you chat with Google’s Gemini AI directly from your command line.

But it is more than just a chatbot. It is an open-source AI agent powered by Gemini.

Why Use the CLI? (Browser vs. Terminal)

Working from the browser is great for general questions, but for building software, the Command Line Interface (CLI) is superior. Here is why:

Direct File System Access: You don't need to copy-paste code. You can simply say, "Refactor the utils.py file," and the CLI can read it and propose changes directly.
Context Awareness: The CLI knows where it is. It understands your current working directory and the files inside it.
Automation & Scripting: You can pipe output from other commands into Gemini. For example: cat logs.txt | gemini "Find the critical error".
Built-in Tools: Gemini CLI comes battery-included with tools such as:
- File System: Read, write, and list files.
- Web: Fetch URLs and search Google (grounding your code in real-time info).
- Shell: Execute terminal commands safely.
Customization & Extensions: Because it is open source, you can extend it. It supports the Model Context Protocol (MCP), allowing you to build custom tools that connect to your database or internal APIs.

Real-Life Examples and Ideas to Build

Here are a few ways you can start using Gemini CLI today to boost your productivity.

1. The "Smart" Code Reviewer

Instead of manually checking your code for style issues, use Gemini CLI to review it before you push.

Command:

gemini "Review main.py for potential bugs and clean code practices. Output the suggestions as a list."

Why it works: The CLI reads the file directly and uses Gemini’s logic to find edge cases you might have missed.

2. Instant Data Analyst

Imagine you have a messy CSV file or a log file and you need quick insights.

Scenario: You have a file called server_logs.csv.
Command:

gemini "Read server_logs.csv and tell me which IP address appears the most frequently. Then, use grep to count exactly how many times it appears."

Why it works: The agent combines its ability to write code (to analyze the CSV) with system tools (like grep) to verify the data.

3. Rapid Prototyping

Need to scaffold a new feature?

Command:

gemini "Create a new folder called 'blog-app'. Inside it, create an index.html file with a basic responsive layout and a style.css file with a dark mode theme."

Result: The CLI actually creates the folder and writes the files for you. You are ready to start coding immediately.

How to Get Started: Installation

Ready to try it? Installing Gemini CLI is straightforward. You will need Node.js installed on your machine.

(More details, see the official documentation at geminicli.com)

Step 1: Open your terminal.

Step 2: Run the install command.

npm install -g @google/gemini-cli@latest

Step 3: Authenticate.
You can start the tool by simply typing:

gemini

On your first run, it will ask you to authenticate. You can usually log in directly with your Google account, which gives you a generous free tier to start experimenting.

And once it is installed you can see it running in your terminal:

Step 4: Enable the Latest Features (Crucial Step!)
To use the powerful Gemini 3 Flash model, you need to enable preview features.

Type /settings in the CLI.
Look for "Preview Features".
Set this to "True". This ensures you are running on the latest, fastest, and most capable models available.

Under the Hood: How Gemini CLI Works

You might be wondering, "How does a text box in my terminal know how to run code?"

Gemini CLI uses a concept called the ReAct Loop (Reason + Act). Here is the simple explanation of what happens when you type a command:

Thought: You ask a question (e.g., "Why is my build failing?"). The AI analyzes your request.
Tool Use: Instead of just guessing, the AI sees it has "tools" available—like reading a file or running a shell command. It decides, "I should run the build command to see the error."
Observation: The CLI runs the command, captures the output (the error message), and feeds it back to the AI.
Answer: The AI reads the error and gives you the specific fix.

This "Loop" allows the CLI to act like a real pair programmer rather than just a text generator.

Navigating the Terminal: Essential Commands

Once you have installed Gemini CLI, you aren't just typing into a void. There are specific commands to help you control the agent.

/help: The most important command. Use this to see what the agent can do and how to interact with it.
/settings: Use this to explore your setup. This is where you configure your Assistant settings and change the Mode.
/docs: Opens the documentation if you need to read up on advanced features.

How to Reference Your Files

This is the superpower of the CLI. You don't need to copy-paste code. To ask Gemini about a specific file, simply use the @ symbol.

Example:

"Can you explain the logic inside @main.py and suggest improvements?"

The agent will read that specific file and base its answer on the code inside it.

Understanding Context and `GEMINI.md`

The Magic of `GEMINI.md`

You can teach the CLI about your project by creating a special file called GEMINI.md. Think of this as a "ReadMe for the AI."

If you place a GEMINI.md file in your folder, the CLI automatically reads it to understand your project rules, coding style, or architecture. You can place this file in:

The Global Directory: For rules that apply to all your projects.
The Current Working Folder: For project-specific rules.
Subdirectories: For rules specific to that module.

Managing Memory

You can verify what the AI currently "knows" or add new things to its memory manually using these commands:

Check Context:
```
/memory show
```
This will list everything currently in the agent's context.

Add Context Manually:

/memory add "Remember that we are using Python 3.11 for this project"

Or you can point it to a specific context file:

/memory add path/to/another/gemini.md

Already excited about the topic, this is the intro, we'll explore more about CLI Extensions, uses for Software Development, Data Analysis, MCP and more in the coming blogs!

Conclusion (This first part!)

Give it a try, explore the open source code, and let me know what you build!

More blogs coming up soon! Subscribe and stay tuned!

Lesly

Una corta intro a las Redes Neuronales Artificiales

leslysandra — Thu, 23 Sep 2021 03:51:26 +0000

Hola Mundo! Seguro escuchaste algo de las Redes Neuronales Artificiales RNAs cuando se habla de la Inteligencia Artificial 🤖

En este blog, comentaré un poco sobre qué son las RNAs o ANNs (Artificial Neural Networks), como también un par de recomendaciones para aprender más sobre este tema.

La inspiración 🤩
El cuerpo humano y en especial el cerebro es una gran fuente de inspiración para entender la inteligencia y pues, de alguna forma, modelarla matemáticamente.

En la imagen a continuación, se puede ver que el "cell body" es la parte del procesamiento de información, así como las dendritas representan el ingreso de los datos a ser procesados.

Es interesante entender su representación matemática: los datos de entrada, la sumatoria de estas entradas, la función de activación y la salida de esta neurona, que probablemente se convierte en entrada para la siguiente neurona en la siguiente capa.

La no-linearlidad 🤔
Entre las diferentes técnicas de Machine Learning, podemos citar a la "regresión lineal" que nos ayuda a predecir: "precio de casa", "crecimiento poblacional", entre otros. Sin embargo, varios problemas de la vida real no se resuelven de forma lineal, y es aquí donde esta combinación de entradas y funciones de activación, permiten tener mejoras significativas en las aplicaciones de Machine Learning, hasta lograr entender el contenido de las imágenes y también entender el lenguaje (cómo cuando le hablas a un asistente de voz)

Recomendaciones 🤓
En este blog, no explico los conceptos básicos, pero si los menciono porque es importante conocer los fundamentos de este tema, para disfrutar mejor la práctica y hacernos más cuestionamientos que nos permitan entender el mundo de la IA.

Si estas aprendiendo sobre Redes Neuronales Artificiales, debes conocer ¿qué es? ¿cómo funciona? ¿cómo se lo programa y/o modifica?... los siguientes puntos:

Perceptrón y Multicapas
Funciones de Activación: Sigmoid, RELU, etc
Descenso de Gradiente (Gradient Descent)
Función Costo
Funciones de Optimización para RNAs
Feed forward y Back propagation
Tipos de RNAs

Herramientas para construir las RNAs, sin taaanto código o matemática por detras:

TensorFlow de Google
PyTorch de Facebook
Caffe de BAIR

Tipos de RNAs:

Redes Convolucionales: Convolutional Neural Networks CNN (estas son las de visión 👀)
Redes Recurrentes: Recurrent Neural Networks RNN (estas para el lenguaje 🗣️)
Generative Adversarial Networks GAN (con las que se puede hacer los deepfake 😱)

Reconocimiento de imágenes, como en la imagen 🙈

Si es la primera vez que estas queriendo empezar con las RNAs, te recomiendo seguir el tutorial de TensorFlow con su ejemplo del Fashion MNIST
Además, el curso completo de especialización en Coursera Introduction to TensorFlow for Artificial Intelligence, Machine Learning, and Deep Learning

Hay mucho que hablar sobre las Redes Neuronales Artificiales 🤖, así que nos vemos en otros blogs (o también en mi canal de YouTube)

Python Conference Latam 2021 #PyConLatam

leslysandra — Thu, 23 Sep 2021 02:02:58 +0000

El pasado agosto, se realizó otra versión del PyCon Latam (virtual) donde la comunidad pythonera se hizo presente 🐍. Revisa la agenda completa aquí: https://www.pylatam.org/horario-completo/

El evento de 2 días, organizado en una plataforma muy fácil de acceder, con espacios de discusión (virtual), feria de los auspiciadores, y concursos por participación en las presentaciones y más!

Tuve el honor de compartir una presentación denominada: "Python 🐍 para empezar a aprender Inteligencia Artificial 🤖"

PyCon Latam 2021 🐍 - hablemos de python para IA 🤖

leslysandra — Thu, 09 Sep 2021 04:27:38 +0000

En agosto 2021, volvió el (Python Conference Latam)[https://www.pylatam.org/] 🥳 (virtual), un evento de dos días reuniendo a la comunidad pythonera en charlas de alto nivel, conferencias de otros latinos en tecnología, espacios de discusión, acercamiento a las empresas auspiciadores y sobre todo mucho compartir conocimiento y networking.

Tuve la oportunidad de participar como expositora con la presentación titulada: "Python 🐍 para empezar a aprender Inteligencia Artificial 🤖"

()[https://twitter.com/FerroRodolfo/status/1431705921680379906]

Qué bueno es saber que Python es una herramienta muy poderosa y bastante usada en el mundo de la Inteligencia Artificial.
Repasando primero los conceptos básicos de Inteligencia Artificial y Machine Learning, y explorar las opciones de proyectos para iniciantes en este mundo (practicar es lo mejor, y python es una herramienta increíble para empezar)

Las diapositivas de mi presentación aquí: https://bit.ly/3tqAUO4

DEV Community: leslysandra

Andrew Ng's new open-source project, Context Hub, attempts to solve a problem every API provider has right now whether they know it or not. Coding agents are getting your API wrong.

Context Hub Has 68 APIs. Add Yours.

Gemini 3.1 Pro & "Nano Banana 2": A Technical Review and Real-World Tests for Devs

Gemini 3.1 Pro: Your New Logic Sidekick

The Pain Point it Solves: The Output Limit

Testing the Logic: The ARC-AGI-2 Challenge

2. Nano Banana 2: Fast, Furious, and... Can it Read?

Try It Yourself (Hands-on Tests)

Test A: The UI Mockup (Perfect for Frontend)

Test B: Search Grounding (What wasn't in the dataset)

Extra: One more test

Conclusion: Where Should You Start?

Last part! (3 out of 3) to Get started with Gemini CLI

Part 3: Mastering Gemini CLI – Content Creation, Learning, and Multimodality

leslysandra ・ Feb 9

Part 3: Mastering Gemini CLI – Content Creation, Learning, and Multimodality

1. Content Creation with Extensions (The "NanoBanana" Workflow)

How to Connect NanoBanana

2. Gemini CLI as Your Personal Tutor

Scenario: The University Student / Self-Learner

3. Grounding Your Knowledge with Web Search

Conclusion: The "All-in-One" Developer

How to get started with Gemini CLI :)

How to Get Started and Build with Gemini CLI (Powered by Gemini 3 Flash)

leslysandra ・ Feb 9

Part 2: Supercharging Your Workflow with Gemini CLI

1. The Power of MCP (Model Context Protocol)

2. Using Extensions: The Google Docs Example

3. Software Development: The GitHub Workflow

4. Data Analysis on the Fly

Final Thoughts

How to Get Started and Build with Gemini CLI (Powered by Gemini 3 Flash)

What is Gemini CLI?

Under the Hood: How Gemini CLI Works

Why Use the CLI? (Browser vs. Terminal)

Real-Life Examples and Ideas to Build

1. The "Smart" Code Reviewer

2. Instant Data Analyst

3. Rapid Prototyping

How to Get Started: Installation

Conclusion

How to Get Started and Build with Gemini CLI (Powered by Gemini 3 Flash)

What is Gemini CLI?

Why Use the CLI? (Browser vs. Terminal)

Real-Life Examples and Ideas to Build

1. The "Smart" Code Reviewer

2. Instant Data Analyst

3. Rapid Prototyping

How to Get Started: Installation

Under the Hood: How Gemini CLI Works

Navigating the Terminal: Essential Commands

How to Reference Your Files

Understanding Context and GEMINI.md

The Magic of GEMINI.md

Managing Memory

Conclusion (This first part!)

How to Get Started and Build with Gemini CLI (Powered by Gemini 3 Flash)

What is Gemini CLI?

Why Use the CLI? (Browser vs. Terminal)

Real-Life Examples and Ideas to Build

1. The "Smart" Code Reviewer

2. Instant Data Analyst

3. Rapid Prototyping

How to Get Started: Installation

Under the Hood: How Gemini CLI Works

Navigating the Terminal: Essential Commands

How to Reference Your Files

Understanding Context and GEMINI.md

The Magic of GEMINI.md

Managing Memory

Conclusion (This first part!)

More blogs coming up soon! Subscribe and stay tuned!

Una corta intro a las Redes Neuronales Artificiales

Python Conference Latam 2021 #PyConLatam

PyCon Latam 2021 🐍 - hablemos de python para IA 🤖

Understanding Context and `GEMINI.md`

The Magic of `GEMINI.md`

Understanding Context and `GEMINI.md`

The Magic of `GEMINI.md`