Optimize bloated JSONL session files by truncating large tool results while preserving conversation continuity.

Transform 2MB+ session files into lean, fast-loading versions without losing context. Built for Claude Code, Clawdbot sessions, and any JSONL-based chat logs.

# Analyze your sessions
python analyze-sessions.py ~/.claude/projects/

# Preview changes (safe, no modifications)
python truncate-sessions.py ~/.claude/projects/ --dry-run

# Apply truncation with automatic backups
python truncate-sessions.py ~/.claude/projects/ --confirm

cd web-ui
npm install && npm run dev
# Open http://localhost:3000

Drag & drop your JSONL files, analyze bloat, download optimized versions. Perfect for one-off files or non-technical users.

Claude Code session files balloon to 2MB+ because tool results contain massive outputs - file contents, logs, command output. This kills context loading speed and wastes tokens.

Surgical truncation: Keep first 2.5KB + last 2.5KB of large tool results, insert a size marker. Preserves conversation flow, cuts 90% of bloat.

Before (bloated tool result):

{"role":"tool","tool_result":"File contents:\nline 1\nline 2\n...50,000 lines of logs..."}

After (truncated):

{"role":"tool","tool_result":"File contents:\nline 1\nline 2\n\n[TRUNCATED: 197KB removed, keeping first + last 2.5KB each]\n\n...line 49,999\nline 50,000"}

• Parse each JSONL line and identify tool results over threshold (default 5KB)
• Truncate large results: keep first 2.5KB + last 2.5KB, insert size marker
• Preserve all user messages, assistant responses, and conversation metadata

• Large tool results over 5KB (configurable via --threshold)
• Preserves first and last portions with size marker
• Never touches user messages, assistant responses, or metadata

• Atomic backups - Timestamped backups before any modification
• Dry-run mode - Preview all changes before applying
• JSON validation - Validates output before writing
• Rollback ready - Restore from .backup_* files anytime

CLI Tools:

• Python 3.9+ (no external dependencies)

Web Interface:

• Node.js 18+
• npm or yarn

MIT