Show HN: Replay an AI Coding Session and Inspect Its Requests

Category: ai-ml

Tags: llm, visualization, ai-agents, web-app, debugging

Score: 6.3/10 (Innovation: 6, Technical: 7, Documentation: 7, Utility: 5)

LLM Visualizer is a Vite/React app that provides two main visualizations: attention-based token contribution text and a replay of a coding agent session showing requests, tool calls, and costs. It offers insights into how agent harnesses like Claude Code work, including prompt caching and cost breakdowns. The project is interesting for its educational value and technical depth in visualizing LLM internals.

Target audience: AI developers, ML engineers, and technical users working with LLM agents

Repository: https://ishamf.dev/p/agent-harness-replay/ · TypeScript · MIT · 17 stars

View on Hacker News