What This Does

This playbook takes your exported AI chat history (Claude, ChatGPT, or any other), breaks it into overlapping chunks, summarizes each chunk, and compiles everything into a structured mind map. The output works as a standalone knowledge document or as input for tools like NotebookLM. Inspired by a Reddit user who needed to import massive chat histories into NotebookLM but the JSON was too large — so they chunked, summarized, and created a mind map that fit perfectly.

Prerequisites

• Claude Code installed and configured
• Exported chat history (JSON, TXT, or any text format)
• Optional: NotebookLM or Obsidian for viewing the mind map

The CLAUDE.md Template

Copy this into a CLAUDE.md file in your chat processing folder:

# Chat History Mind Mapper

## Goal
Process exported AI chat histories into structured summaries and mind maps. Break large files into overlapping chunks, summarize each, then compile into a navigable knowledge map.

## Directory Structure
- `exports/` — Raw exported chat files (JSON, TXT)
- `chunks/` — Split conversation chunks with overlap
- `summaries/` — Per-chunk summaries with topic tags
- `output/` — Final mind map and compiled summary

## Processing Pipeline

### Phase 1: Chunk
Split the exported chat file into overlapping chunks:
- Target size: ~30,000 words per chunk
- Overlap: Each chunk shares ~5,000 words with the next
- Pattern: If the file is ABCD, chunks are AB, BC, CD
- Name chunks sequentially: chunk-01.md, chunk-02.md, etc.
- Preserve conversation structure (don't split mid-message)

### Phase 2: Summarize
For each chunk, generate a summary:
- 500-1,000 word summary of key topics discussed
- List of specific topics, questions asked, and insights gained
- Tag which chunk number the topic came from (for back-reference)
- Identify recurring themes across chunks

### Phase 3: Compile
Merge all summaries into a single document:
- Group topics by theme (not by chunk order)
- Create a hierarchical mind map structure
- Include chunk references so topics can be traced to original conversations
- Add a "Key Insights" section with the most valuable discoveries
- Generate a topic index for quick navigation

## Chunking Rules
1. Never split in the middle of a conversation turn
2. Overlap ensures no context is lost at chunk boundaries
3. Chunks should be roughly equal in size
4. For JSON exports, flatten to plain text before chunking

## Summary Format
For each chunk summary:

Chunk [N] Summary

Source: chunk-[N].md (words X,XXX - X,XXX of original)

Topics Covered

• Topic 1: Brief description [chunk-N]
• Topic 2: Brief description [chunk-N]

Key Questions Asked

• Question 1?
• Question 2?

Insights

• Insight 1
• Insight 2

Recurring Themes

• Theme that appears in multiple chunks

## Mind Map Format (output/mind-map.md)

Chat History Mind Map

Theme 1: [Theme Name]

• Sub-topic A [chunk-2, chunk-5]
  • Key insight
  • Related questions explored
• Sub-topic B [chunk-3]
  • Key insight

Theme 2: [Theme Name]

...

Key Insights

1. Most important discovery [source chunks]
2. Second insight [source chunks]

Topic Index

Topic	Chunks	Summary
Topic A	1, 3, 7	Brief description

## Commands
- "/process [filename]" — Run the full pipeline on an export file
- "/chunk [filename]" — Run Phase 1 only
- "/summarize" — Run Phase 2 on existing chunks
- "/compile" — Run Phase 3: generate mind map from summaries
- "/search [topic]" — Find all mentions of a topic across chunks
- "/themes" — Show recurring themes across all conversations

Step-by-Step Setup

Step 1: Create the project structure

mkdir -p ~/chat-mapper/{exports,chunks,summaries,output}
cd ~/chat-mapper

Step 2: Export your chat history

Claude: Go to Settings → Export Data. You'll receive a JSON file.

ChatGPT: Go to Settings → Data Controls → Export Data.

Place the exported file in the exports/ folder.

Step 3: Save CLAUDE.md and process

cd ~/chat-mapper
claude

Try: "/process exports/conversations.json"

Example Usage

Process a full export:

"/process exports/claude-chats.json — Break this into overlapping chunks of ~30K words, summarize each, and create a mind map of all topics I've discussed with Claude."

Just chunk a large file:

"/chunk exports/chatgpt-export.json — Split this into overlapping chunks. The file is in reverse chronological order."

Search for a specific topic:

"/search machine learning — Find every conversation where I discussed machine learning and show me the key questions I asked and insights I gained."

Find patterns in your conversations:

"/themes — What topics do I keep coming back to? What questions do I ask repeatedly? What themes connect my different conversations?"

Prepare for NotebookLM:

"/compile — Generate a single compiled summary document under 50,000 words that I can upload as a source to NotebookLM."

Tips

• Overlapping chunks prevent information loss: The overlap pattern (AB, BC, CD) ensures that topics spanning chunk boundaries are captured in at least one complete chunk.
• Tag chunk numbers in summaries: When the mind map says "discussed in chunk-3 and chunk-7," you can go back to those specific chunks to read the full conversation. This traceability is the most valuable feature.
• Use NotebookLM for the final output: The compiled summary fits well as a NotebookLM source. You can then ask NotebookLM questions about your entire chat history.
• Process periodically: Export and process your chats quarterly. This builds a personal knowledge archive of everything you've discussed with AI.
• Multiple AI histories: You can process exports from Claude, ChatGPT, and other tools into the same mind map for a unified view.

Troubleshooting

Problem: JSON export is too complex to parse

Solution: Ask Claude to flatten the JSON into plain text first: "Read exports/conversations.json and convert it to a plain text file with each message on its own line, preserving the role (user/assistant) and timestamp."

Problem: Summaries miss important details

Solution: Reduce chunk size (from 30K to 15K words) for more detailed summaries. Smaller chunks mean more granular analysis.

Problem: Mind map is too large or unfocused

Solution: Add a focus filter: "Only include topics related to [your interest area] in the mind map." This filters out casual conversations.