Skip to content
ATAI Today Brief
HomeNewsConceptsGuidesToolbox
AboutSubscribeUA
Subscribe

AI Today Brief

The daily AI-engineering brief. Built in public. EN · UA.

XTelegramLinkedInYouTubeRSS
NewsConceptsGuidesSubscribeAdvertiseAboutEditorial policyAI disclosurePrivacyTerms

© 2026 AI Today Brief. All rights reserved.

  1. Home/
  2. News/
  3. Agents & MCP/
  4. Mindwalk: Replay coding-agent sessions on a local 3D codebase map
Agents & MCP

Mindwalk: Replay coding-agent sessions on a local 3D codebase map

July 12, 2026· 5 min read
OKCurated by Oleksandr Kuzmenko, AI Product Engineer·Updated July 12, 2026·Sources cited on every story
AI-assisted · editor-reviewed·How we use AI
Mindwalk: Replay coding-agent sessions on a local 3D codebase map

Mindwalk is a local Go-based tool that visualizes Claude Code and Codex session logs as interactive 3D maps. It helps developers track agent navigation, edits, and errors directly in their local repository structure.

Impact: Medium

Why it matters

You can audit how your coding agents explore, read, and edit your files to debug context inflation and find logic gaps.

TL;DR

  • 01Mindwalk maps file modifications as bright thermal zones, showing exactly where coding agents read or edited.
  • 02The tool runs entirely locally as a Go binary, protecting sensitive repository structures from being leaked.
  • 03Visual timelines highlight structural events like subagent forks and system prompt compacting.

Key facts

Supported Agent Formats
Claude Code, Codex
Core Tech Stack
Go server, React, Three.js
Privacy Model
100% Local

Local Agent Auditing

Mindwalk parses session logs from Claude Code and Codex completely locally, ensuring no sensitive codebase metadata or telemetry leaves your machine. It translates the raw JSONL traces into interactive Three.js 3D representations (radial trees or treemaps) where file nodes glow based on how deeply and frequently they were accessed during the agentic execution.

Color-Coded Execution Path

Files in the 3D map are styled by their deepest level of interaction: unvisited files remain dark, while touched files display as moss green (seen), moon white (read), or warm amber (edited). The HUD tracks friction signals including mutation phases, context compactions, subagent launches, error rates, and file churn, enabling developers to visually grasp the agent's reasoning footprint.

Command Line Interface

Run the installer script or build the binary directly from source using make build. Launch the server using mindwalk serve, which automatically scans default directories or specified overrides, and spins up the local web client on a random port.

Try it in 2 minutes

curl -fsSL https://raw.githubusercontent.com/cosmtrek/mindwalk/master/scripts/install.sh | sh
mindwalk serve

bash

✓ When to use

  • When analyzing long, complex Claude Code or Codex sessions in large codebases.
  • When debugging agent hallucinations or context-window inflation issues.
  • When you want to review agent changes visually rather than reading raw JSONL logs.

✕ When NOT to use

  • For agents that do not output standard Claude Code or Codex JSONL session formats.
  • When working on closed machines without Go or Node runtime access.

What to do today

  • →Run the quick-install script to set up Mindwalk locally.
  • →Launch the serve command to scan your existing Claude Code session logs.
  • →Navigate the 3D model using the keyboard controls (Space, Arrow keys, E, X) to analyze recent edits.

What the community says

  • “This can help you evaluate agentic capabilites of various llm's. How they explore codebase.”

    — freakynit on Hacker News

  • “If you have a big codebase, text diff is not faster to see what was done here and there, then a visual overview.”

    — lukan on Hacker News

#Claude Code#Codex#Mindwalk#Go#Three.js#React

Sources

  • Mindwalk GitHub Repository
  • HN Comment by freakynit
  • HN Comment by lukan
ShareShare on XShare on LinkedIn
← Previous storyBuilding a Self-Evolving AI Agent in Under One Hundred Lines of LispNext story →Stop Deflecting Complex Technical Questions with 'Ask Claude' and Share Lived Experience

Related stories

  • Agents & MCPDesktop Commander Model Context Protocol Server Grants Claude Terminal and File Controls
  • Agents & MCPExploring System Prompts of Claude, ChatGPT, Gemini, and Specialized AI Agents
  • Agents & MCPGrok 4.5 Tops Perplexity WANDR Orchestrator Benchmark at Half the Cost of Opus
  • Agents & MCPGhostcommit: Exposing Secret-Stealing Prompt Injection in Images

Email digest

Get the morning AI brief

One email a day — the stories that matter for engineers, founders and tech leads. Human-edited, with links to primary sources.

  • ✓120+ sources scanned daily
  • ✓Edited by a human
  • ✓1 email per day
  • ✓EN + UA

By subscribing you agree to the privacy policy.