AgenticLogger -

Inspiration

In the fast-paced world of AI-assisted development, critical context often vanishes into the "scroll-back" abyss. We wanted to create a persistent, git-native memory for every decision, prompt, and cost incurred during a session.

What it does

It acts as an automated tracker. For every interaction, the agent switches branches, generates a dedicated Markdown log, updates a master index, commits the changes to a dedicated logAgent/traces branch, and returns to the working branch—all in one go.

How we built it

Using the GitLab Duo ecosystem and a custom Agentic Prompt definition, we established a strict state-machine protocol. The agent calculates token usage and costs in real-time while managing file I/O and Git operations programmatically.

Challenges we ran into

Ensuring seamless branch switching without disrupting the user's current working state was the biggest hurdle. We also had to refine the prompt-to-cost calculation to provide immediate financial transparency for every generation.

Accomplishments that we're proud of

We successfully build an history of a Gitlab Duo Chat documentation. The project effectively manages its own history, even documenting the creation of its own logo and thumbnails within the same sequence it was defining itself.

What we learned

State management is king. By giving the AI a strict set of "Agentic Constraints," we turned a simple chatbot into a reliable data-entry and version-control powerhouse.

What's next for

GitLab Duo: Agentic Logger will soon be a part of a prompt management solution, tracking versioning prompts, witch will include Agents

Built With

• gilabduo