Documentation

🚀 Getting Started

About Bert

The idea of Bert was actually concived during 2022, During this time, AI assistants were just raising to what they are today, Bert idea begins when the idea of a AI model for you and only you that can be whatever you want it to be, when this Idea raised, Amphydia wasn't yet created, and I, Matias Nisperuza was younger and had no chance on making bert, but times change, for good.

Bert Identity

We frame bert as a service fundamentally oriented around human satisfaction, not raw capability benchmarks or abstract performance metrics.
At its core, bert is built to feel like a reliable presence—a system that users can return to consistently, trust implicitly, and engage with naturally. In practical terms, this means prioritizing clarity over cleverness, helpfulness over verbosity, and dependability over spectacle.
bert is not positioned as an omniscient authority or a cold computational engine. Instead, it operates as a dependable companion:
a system that understands context, respects user intent, and responds in a way that feels grounded, supportive, and predictable.

Installation

# Strongly Recommended- Using PyPI pip install
pip install bert-cli

#Run bert after install
bert

#cloning repo in github - not recommended
git clone https://github.com/mnisperuza/bert-cli.git
cd bert-cli

# Install dependencies
pip install -e .

# Run Bert
bert 

By installing bert you are agreeing to the terms of use and privacy.

First Run

When you first run Bert, you'll see an animated banner followed by a quantization picker. Choose based on your GPU:

• ⭐ INT4-FAST — Best for most users. Works with 4GB+ VRAM.
• 🔷 INT8-BALANCED — Higher quality. Needs 6GB+ VRAM.
• 💎 FP16-HIGH-END — Best quality. Needs 8GB+ VRAM.
• 🖥️ FP32-CPU — CPU mode. Slower but no GPU needed.

⌨️ Commands Reference

Model Commands

Quantization Commands

System Commands

During Generation

🤖 Model Guide

Bert comes with 6 specialized models, each optimized for different tasks.

Which Model Should I Use?

• 💬 Casual chat, quick questions: Bert Nano
• 📝 Writing, summarizing, general tasks: Bert Mini or Main
• 🧠 Complex reasoning, math, analysis: Bert Main (with /*think) or Max
• 💻 Simple code, scripts, debugging: Bert Coder
• 🏗️ Large codebases, architecture: Bert Max-Coder

🧠 Thinking Mode-BETA

Bert Main (Qwen3-1.7B) supports a special thinking mode that shows the model's reasoning process after the response generation We are improving this feature, so expect a major update in a couple of months.

How to Use

/*think - What is the derivative of x³ + 2x²?

The response will show:

1. The model's answer (streamed normally)
2. A thinking box showing the reasoning process
3. Token count (only counts the response, not thinking)

When to Use Thinking

• ✓ Math problems and calculations
• ✓ Logic puzzles and reasoning tasks
• ✓ Complex questions with multiple steps
• ✓ When you want to understand HOW Bert reached an answer

📁 File References-BETA

You can reference files directly in your queries using the @ symbol. Bert can easily find, check and read files in the current directory, although, Features like reviewing files or adding a relative path are still in early development. if you enconter a issue, dont doubt emailing us at mnisperuza1102@gmail.com

Usage

# Reference a file
Check @main.py for bugs

# Multiple files
Compare @old_version.py and @new_version.py

# Relative paths
Review the code in @src/utils/helpers.js

Supported File Types

🎟️ Token System

Bert uses a weekly token system to manage usage. Every week, you get 20,000 free tokens.

Getting a Token

1. Visit the Bert CLI homepage
2. Enter your email
3. Receive your token (format: BERT-XXXX-XXXX-XXXX-XXXX)
4. In Bert, type: /*token YOUR-TOKEN-HERE

Token Commands

# Set your token
/*token BERT-A1B2-C3D4-E5F6-0123

# Check remaining tokens
/*tokens

How Tokens Are Counted

• 📊 Each response uses tokens based on length
• 🧠 Thinking content does NOT count against your tokens
• 🔄 Tokens reset every week (Sunday midnight)
• 📧 One token per email per week

⚙️ Quantization Guide

Quantization reduces model size and memory usage, allowing larger models to run on smaller GPUs.

✨ Tips & Best Practices

Getting Better Responses

• 1. Be specific: "Write a Python function that sorts a list" is better than "Write code"
• 2. Provide context: Share relevant background information
• 3. Use the right model: Bert Coder for code, Bert Main for reasoning
• 4. Reference files: Use @filename instead of pasting code

Keyboard Shortcuts

• ⎋ ESC: Stop generation instantly
• ↑ Up Arrow: Previous command (terminal dependent)
• ⌃C Ctrl+C: Interrupt / stop

Troubleshooting