Agent Mode: Your AI That Can Think and Act

What is Agent Mode?

Agent Mode transforms MyLLM from a simple chatbot into an autonomous AI assistant. Instead of just answering questions, the agent can break complex tasks into steps, reason through each one, and execute actions — all running locally on your device.

Think of it like having a junior developer or personal assistant in your pocket, except it never sends your data to the cloud.

How It Works

When you give the agent a task, it follows a structured thinking process:

• Understanding — The agent parses your request and identifies what needs to be done
• Planning — It breaks the task into manageable steps
• Execution — It works through each step, using tools as needed
• Verification — It checks its work and reports back

Chain-of-Thought Reasoning

The agent uses chain-of-thought (CoT) reasoning, meaning it "thinks out loud" before answering. You can see its reasoning process in real-time — the thinking steps, tool calls, and intermediate results are all visible.

This transparency is a key advantage over cloud AI: you can see exactly how the agent arrived at its answer.

Real-World Examples

Example 1: Expense Analysis

You: "Analyze my expenses.csv and create a summary"

Agent thinks:

• Read the CSV file using File Manager
• Parse the transaction records
• Categorize expenses by type (food, transport, bills, etc.)
• Calculate totals per category
• Generate a formatted summary

Example 2: Code Review

You: "Review this Python function and suggest improvements"

Agent thinks:

• Parse the code structure
• Check for common anti-patterns
• Analyze time/space complexity
• Suggest optimizations with explanations

Example 3: Research Assistant

You: "Help me understand the pros and cons of solar panels for my home"

Agent thinks:

• List key factors to consider
• Research cost estimates
• Analyze environmental benefits
• Consider location-specific factors
• Provide a balanced summary

What Tools Can the Agent Use?

The agent has access to all 20+ built-in tools:

• Code Interpreter — Run and debug code snippets
• File Manager — Read, analyze, and create files
• Calculator — Precise mathematical computations
• Writer — Draft emails, essays, reports
• Summarizer — Condense long content into key points
• Reasoner — Deep logical analysis and problem-solving
• Translator — Translate text between languages

Privacy in Agent Mode

Everything the agent does happens on your device. File reading, code execution, calculations — all local. The agent has no ability to make network requests during its reasoning process.

Getting the Best Results

• Be specific — "Analyze my budget spreadsheet and find areas to cut spending" works better than "help with money"
• Use larger models — Agent Mode benefits from models with better reasoning. We recommend Qwen 3.5 4B or 8B.
• Be patient — Multi-step tasks take longer than simple Q&A. The agent is thinking through each step.
• Check the reasoning — Read through the agent's thinking steps. If it's going in the wrong direction, you can redirect it.

Try Agent Mode →