screen page

About the Project

AitMate was inspired by a simple idea: what if our computer had a personal AI assistant that could do more than just answer questions?

We wanted to build an AI agent that can understand natural-language instructions, interact with desktop applications and websites, summarize information, prepare responses, and execute tasks on the user's behalf — while keeping the user in control of every sensitive action.

The main challenge was finding the right balance between autonomy and security. An agent that can access a computer and perform actions can be extremely useful, but giving an AI unrestricted access to a user's system is not acceptable. That's why AitMate was designed around a permission-first architecture.

AitMate can analyze a requested task, decide which tools are needed, execute safe actions, and ask for explicit approval before performing sensitive operations such as sending messages, modifying important files, or executing terminal commands.

We also wanted the assistant to work efficiently without requiring a powerful GPU running continuously. The architecture therefore separates lightweight local operations from more demanding AI reasoning and uses an internet connection when external services or cloud-based intelligence are required.

For the hackathon, we are also exploring how AitMate can connect with GitLab's agentic development ecosystem. This allows the assistant to participate in software-development workflows such as analyzing issues, working with repositories, reviewing code, running tests, identifying security problems, and assisting with fixes and deployments.

What We Learned

Building AitMate taught us that creating an autonomous AI agent is not only about making a model smarter. The real challenge is designing a reliable system around the model:

Tool and permission management
Human-in-the-loop approval
Secure command execution
Browser and desktop automation
Agent planning and task execution
Context and memory management
Error handling and recovery
Integration with external services
Efficient local/cloud AI orchestration

Challenges

One of our biggest challenges was designing an agent that is useful without becoming uncontrolled. Every action needs to have clear permissions, and potentially dangerous operations must require user approval.

Another challenge was making the system lightweight enough to run as a desktop assistant without constantly consuming large amounts of CPU, memory, or GPU resources.

We also had to think about how the agent should recover when a task fails. Instead of simply reporting an error, AitMate is designed to analyze failures, explain what happened, and, when permitted, attempt a correction.

Ultimately, AitMate is our attempt to build a different kind of AI assistant: not just an AI that talks to you, but an AI that can work for you — while you remain in control.