claude-devtools

Product Introduction

1. Definition: claude-devtools is an open-source desktop application for technical session log analysis, specifically designed to reconstruct and visualize Claude Code AI agent activities. It operates as a local log parser, processing raw session data stored in ~/.claude/ directories.
2. Core Value Proposition: It solves Claude Code’s terminal opacity by transforming hidden JSON logs into structured, visual timelines—revealing every file access, tool execution, token allocation, and context compaction event without modifying Claude’s operations or requiring API keys.

Main Features

1. Context Reconstruction: Tracks per-turn token attribution across 7 categories (CLAUDE.md, skill activations, tool I/O, etc.). Visualizes context window filling/compaction via dynamic graphs, showing exact state changes during AI reasoning cycles.
2. Tool Call Inspector: Expands collapsed actions (e.g., "Read 3 files") into syntax-highlighted file paths, inline Edit diffs, Bash command outputs, and subagent execution trees—rendering every granular operation.
3. Team & Subagent Trees: Isolates nested agent workflows into recursive execution trees with dedicated metrics (token cost, duration, errors), enabling hierarchical debugging of multi-agent collaborations.
4. Notification Triggers: Detects critical events via regex-based alerts for .env access, token thresholds, or sensitive paths (e.g., /payment/). Custom triggers map to color-coded system notifications for real-time monitoring.
5. SSH Remote Sessions: Securely inspects logs on remote machines using SSH (supports ~/.ssh/config, agent forwarding). Enables instant switching between local/remote sessions without data syncing.
6. Command Palette: Cmd+K enables cross-session keyword search, returning context snippets with highlighted matches for direct navigation to specific messages or tool calls.
7. Multi-Pane Layout: Supports side-by-side session comparisons with drag-and-drop tabs, split views, and synchronized scrolling for parallel workflow analysis.

Problems Solved

1. Pain Point: Claude Code’s CLI obscures critical details (e.g., which files were edited, actual token usage per step), forcing reliance on noisy --verbose JSON dumps. claude-devtools eliminates guesswork with structured, queryable visuals.
2. Target Audience: AI engineers debugging agent logic; security teams auditing file/tool access; cost-optimization specialists analyzing token efficiency; DevOps managing remote Claude deployments.
3. Use Cases: Auditing accidental .env exposure; optimizing context usage after compaction; comparing subagent performance across sessions; replicating bugs via exact tool-call sequences.

Unique Advantages

1. Differentiation: Unlike GUI wrappers that intercept API calls, claude-devtools operates passively—analyzing existing logs with zero Claude Code interference. Outperforms manual log parsing with compression-aware context visualization and subagent tree mapping.
2. Key Innovation: Reverse-engineers Claude’s internal state transitions using log metadata, reconstructing ephemeral context windows and compaction events otherwise irrecoverable from terminal output.

Frequently Asked Questions (FAQ)

1. Does claude-devtools require Claude Code configuration changes?
   No. It reads native session logs from ~/.claude/ without modifying Claude Code’s behavior, API connections, or prompts.
2. How does claude-devtools handle data privacy for sensitive logs?
   All processing occurs locally; zero data leaves your machine. SSH remote sessions use existing credentials without storing log data externally.
3. Can I analyze historical Claude sessions with claude-devtools?
   Yes. It parses all existing logs in ~/.claude/, enabling retroactive inspection of any past session—including token costs and tool-call sequences.
4. What OS platforms support claude-devtools?
   Native builds for macOS (Intel/Apple Silicon), Windows 10+, and Linux (AppImage/.deb/.rpm), plus Docker for browser-based access via localhost:3456.