Mindwalk Replays Your Coding-Agent Sessions on a 3D Map of Your Codebase

Direct Answer

Mindwalk is a new open-source tool (Go + TypeScript, v0.1.0) that replays Claude Code and Codex session logs as light moving across a 3D “city map” of your repository. It shows which files the agent searched, read, and edited — turning a wall of JSONL into a visual shape you can grasp at a glance. It runs fully local; no session data leaves your machine.

A session log answers what an agent did. Mindwalk tries to answer how the agent understood the task: which parts of the repo it treated as relevant, where it explored before acting, and whether its footprint matched the scope you had in mind.

Why a 3D map instead of the raw log

Reading session JSONL line by line tells you the agent called a tool, then another tool. It does not tell you whether the agent’s attention matched the problem. Mindwalk’s bet is that attention is a shape, and shapes are faster to judge than text.

The repo is drawn as a night map. As the session plays back, the files the agent touched glow — everything else stays dark. You see the agent’s mental model as a footprint.

How it works

Two artifacts, kept deliberately separate:

Trace — the session log normalized into an ordered stream of file-touch events. One adapter per agent format (Claude Code and Codex today), so adding a new agent is an adapter, not a rewrite.
Citymap — a deterministic layout of the repository. The same tree always produces the same map, so replays are comparable across sessions.

A local Go server joins the two and serves a React/Three.js frontend. There is no cloud, no upload, no account.

Reading the picture

Mindwalk gives you several lenses on the same run:

Tree / Terrain views — the repo as a radial tree or a treemap plain. Glow is proportional to how deeply and how often a file was touched.
Touch states — each file keeps its deepest touch: seen (moss green), read (moon white), edited (warm amber), unvisited (dark). A HUD folds friction signals — error rate, churned files, edits after the last verify — into a review strip.
Playback deck — scrub or play the session over a bucketed histogram. Bars sit on a cool/warm spectrum: observation stays cool (search, read, exec), mutation glows warm (edit, verify), so editing phases jump out.
Timeline marks — context compactions, subagent launches, and user turns are click-to-jump targets.

Commands

mindwalk serve [--port N] [--no-open] [--claude-dir DIR] [--codex-dir DIR]
mindwalk open [--no-open] <session.jsonl>   # open one specific session
mindwalk build <repo> [-o out]              # write the repository citymap JSON
mindwalk trace <session> [-o out]           # write the normalized trace JSON

With no arguments, mindwalk scans ~/.claude/projects and ~/.codex/sessions, serves the UI on a random local port, and opens a browser. Windows archives are on GitHub Releases; or build from source with make setup && make build.

Who it is for

Mindwalk is most useful for the moments when an agent surprises you: a refactor that touched 40 files, a session that wandered before it found the bug, a subagent launch you did not expect. Instead of replaying the log in your head, you watch the footprint.

For context on how these two agents compare head to head, see our breakdown of Claude Code vs Codex in the terminal battle, and the broader landscape in the Complete Guide to AI Coding Agents in 2026.

FAQ

Does Mindwalk send my session data anywhere? No. Mindwalk is a local Go binary that reads Claude Code and Codex session logs on your machine and serves the UI from a local port. The README explicitly states no session data leaves your machine.

Which agents does Mindwalk support? As of v0.1.0, Mindwalk ships adapters for Claude Code (~/.claude/projects) and OpenAI Codex (~/.codex/sessions). The architecture uses one adapter per agent format, so additional agents can be added without rewriting the core.

Is Mindwalk ready for daily use? It is early: the repository was created in July 2026, has one contributor, and is at v0.1.0. It is a useful inspection tool for anyone reviewing or learning from agent sessions, but treat it as a young project rather than a finished product.

Ever wonder if Claude gives a better answer than GPT for a specific coding problem? aiFiesta lets you compare 9+ premium models side-by-side in one chat. No more juggling tabs.