AI-Assisted Developer Tool
ZZYAgent: Local Coding Assistant
A personal prototype for learning how a coding assistant connects model responses to actions in a local project. The updated version shares one runtime across its command line and terminal interface, and adds saved sessions, structured plans, context budgeting, and checks before reporting a task as complete.
Period
Jun - Jul 2026
Tools / Tech
PythonDeepSeekTyperRichTool CallingPytest
Why I built it
I wanted to follow the whole loop: how an assistant reads code, chooses a tool, reacts to a failed test, and decides it has enough evidence to stop.
Links
What it includes
- Connects file reading and editing, code search, shell commands, tests, and Git diffs through a tool-calling loop.
- Saves and resumes local sessions, limits long tool outputs, and summarizes older context.
- Tracks a task plan and checks for diff review and validation after edits.
What I worked on
- Defined the prototype's workflow and used AI assistance to organize its runtime, tools, terminal interface, and tests.
- Reviewed repeated calls, interrupted sessions, and failed commands to make the next action easier to understand.
- Kept build notes and small examples that explain how the agent's pieces fit together.
What I Learned
- Developed a practical understanding of tool calls, context limits, and session state.
- Learned why running a test and inspecting its result are different from asking a model whether a change is correct.