One turn, end to end.
Watch a user message become model input, tool execution, session events, reducer state, and terminal output.
Open walkthroughWalkthroughs
Follow one agent idea until the boundary becomes visible.
These are guided traces through tinyloop's code. They are not production recipes and they are not generic agent lessons. Each walkthrough keeps one small mechanism in view long enough to make it understandable.
Start here
Two focused paths through the codebase.
The first follows one user turn from input to rendered transcript. The second adds a tiny read-only tool so you can see the model/tool contract without adding much power.
One tiny tool.
Add a small list_files tool and trace the schema, handler, registry entry, and UI event details.