Show HN: Mindwalk – Replay coding-agent sessions on a 3D map of your codebase
Mindwalk is a novel visualization tool that transforms raw coding-agent session logs into interactive 3D maps of your codebase, revealing how AI agents perceive and interact with your project. It addresses the black-box problem of agent actions, allowing developers to see an agent's "understanding" at a glance. This "Show HN" offers a clever, local solution for debugging and comprehending AI's code exploration, appealing to the developer community's interest in AI transparency.
The Lowdown
Mindwalk is an innovative visualization tool presented as a "Show HN" that aims to demystify the internal workings of AI coding agents by replaying their sessions on an interactive 3D map of a codebase. It tackles the fundamental challenge of understanding an agent's decision-making process, exploration patterns, and overall "footprint" within a repository, which raw session logs fail to convey. By transforming opaque JSONL logs into a visual narrative, Mindwalk offers developers unprecedented insight into how AI interacts with their code.
- The Problem: Traditional agent session logs only record actions, making it difficult to discern which parts of a repository an agent considered relevant, where it explored, or if its scope matched developer expectations.
- The Idea: Mindwalk visualizes the repository as a "night map," where agent activity (search, read, edit) causes parts of the map to glow, making the agent's understanding immediately apparent.
- Key Features: It's a fully local Go binary supporting Claude Code and Codex session logs, ensuring no data leaves the machine.
- Visualization Modes: Offers "Tree" and "Terrain" views of the repository, where the glow intensity indicates how deeply and frequently a file was touched.
- Interaction and Playback: Files display "touch states" (seen, read, edited) and a HUD shows friction signals. A playback deck allows scrubbing through sessions, with a cool/warm spectrum indicating observation vs. mutation actions.
- Navigation & Inspection: Timeline marks (◇, ○, ▸) highlight key events, serving as click-to-jump targets. An inspector pins file visit history and allows jumping the playhead to specific moments.
- Technical Architecture: Internally, it separates session data into a normalized "trace" and a deterministic "citymap" for consistent layout. A Go server connects these with a React/Three.js frontend.
Mindwalk thus provides a critical lens through which developers can observe, debug, and ultimately better understand the often-obscure processes of AI coding agents, making their contributions to a codebase more transparent and controllable.