VSCode, 
Kilocode, I built this project to explore an AI-assisted coding stack end-to-end and learn how agent workflows translate into token cost. The setup combines three tools, each solving a distinct layer of the problem.
Why this stack?
- VSCode: IDE chosen for better agent compatibility. It exposes the APIs and extension surface that AI-coding agents need to read context, edit files, and run terminal commands reliably.
- Kilocode: Open-source AI agent. Studying its source lets me understand the agent workflow step by step, then customize modes, rules, and tool prompts to fit my own tasks.
- GoModel: Self-hosted LLM API proxy (see ENTERPILOT/GoModel). It intercepts requests to the upstream LLM provider, logs full request/response payloads, and lets me inspect the system prompt, tools, and message history per call.
What I’m optimizing
- Workflow understanding: Compare how Kilocode plans, edits, and validates against how I work manually; reuse the best patterns.
- Token usage: Use GoModel logs to measure per-task prompt size, tool-call overhead, and repeated context. Trim system prompts, cache stable rules, and split large tasks into smaller, focused runs.