Chatgpt Search

chatgpt-search

SQLite FTS5 (SQLite full-text search) engine for ChatGPT conversation exports. BM25-ranked full-text search (relevance scoring) with title boosting, code separation, TF-IDF (term-frequency/inverse-document-frequency) keyword extraction, and filtering by date, role, model, and language.

Setup

cd /path/to/skills/chatgpt-search
./scripts/setup.sh /path/to/your/conversations.json
export PYTHONPATH=/path/to/skills/chatgpt-search/src

• Claude Code: copy this skill folder into .claude/skills/chatgpt-search/
• Codex CLI: append this SKILL.md content to your project's root AGENTS.md

For the full installation walkthrough (prerequisites, verification, troubleshooting), see references/installation-guide.md.

Staying Updated

This skill ships with an UPDATES.md changelog and UPDATE-GUIDE.md for your AI agent.

After installing, tell your agent: "Check UPDATES.md in the chatgpt-search skill for any new features or changes."

When updating, tell your agent: "Read UPDATE-GUIDE.md and apply the latest changes from UPDATES.md."

Follow UPDATE-GUIDE.md so customized local files are diffed before any overwrite.

Repo: ./ Data: <your-export-path>/conversations.json Default DB: ~/.chatgpt-search/index.db

Quick Start

cd . && ./scripts/setup.sh <your-export-path>/conversations.json
export PYTHONPATH=./src
python -m chatgpt_search.cli "your topic query" --limit 10

Decision Tree

Need to search past ChatGPT conversations?
  |
  +-- Know a topic/keyword? --> Full-text search: "query"
  |     +-- Want only user messages? --> add --role user
  |     +-- Want a specific model's responses? --> add --model gpt-5
  |     +-- Want a date range? --> add --since 2025-01 --until 2025-06
  |     +-- Want a specific language? --> add --lang ru
  |
  +-- Know a conversation ID? --> --conversation <id> (or partial ID)
  |
  +-- Want to explore keywords?
  |     +-- Top corpus keywords --> --keywords
  |     +-- Keywords for a conversation --> --keywords --keywords-conversation <id>
  |
  +-- Want corpus overview? --> --stats
  |
  +-- Need to search non-ChatGPT docs? --> Use your project's document search skill
  +-- Need to search Apple Notes/Obsidian? --> Use a dedicated document search tool
  +-- Need web search? --> Use web-search skill (optional companion, not required)

Setup

cd . && ./scripts/setup.sh <your-export-path>/conversations.json

This installs dependencies (scikit-learn, langdetect) and builds the index from the provided conversations.json location. Rebuild takes ~26 seconds on the full corpus (1,514 conversations, 16,689 messages).

CLI Reference

# Set PYTHONPATH (or install the package)
export PYTHONPATH=./src

# --- Search ---

# Full-text search
python -m chatgpt_search.cli "transformer attention"

# Date filtering
python -m chatgpt_search.cli "kubernetes" --since 2025-01
python -m chatgpt_search.cli "pytorch" --since 2025-06 --until 2025-12

# Role filtering (search only user messages or assistant responses)
python -m chatgpt_search.cli "pricing strategy" --role user

# Model filtering (partial match)
python -m chatgpt_search.cli "code review" --model gpt-5
python -m chatgpt_search.cli "reasoning" --model o3

# Language filtering
python -m chatgpt_search.cli "machine learning" --lang en
python -m chatgpt_search.cli "обучение" --lang ru

# Phrase queries (exact match)
python -m chatgpt_search.cli '"attention is all you need"'

# Prefix queries
python -m chatgpt_search.cli "transfor*"

# Limit results
python -m chatgpt_search.cli "topic" --limit 5
python -m chatgpt_search.cli "topic" -n 50

# --- Browse ---

# Browse a full conversation
python -m chatgpt_search.cli --conversation <conversation-id>
python -m chatgpt_search.cli -c <partial-id>

# --- Keyword Exploration ---

# Top keywords across the corpus (by total TF-IDF score)
python -m chatgpt_search.cli --keywords

# Keywords for a specific conversation
python -m chatgpt_search.cli --keywords --keywords-conversation <conversation-id>

# --- Corpus Info ---

# Corpus statistics (conversations, messages, keywords, models, dates)
python -m chatgpt_search.cli --stats

# --- Index Management ---

# Rebuild index (includes TF-IDF enrichment)
python -m chatgpt_search.cli --rebuild --export /path/to/conversations.json

# Custom database location
python -m chatgpt_search.cli --db /path/to/index.db "query"

Search Syntax

FTS5 query syntax (SQLite full-text query operators) is supported:

Architecture

• Engine: SQLite FTS5 (SQLite full-text search) with BM25 ranking (relevance scoring)
• Indexing: Message-level rows, conversation metadata joined at query time
• Boosting: Title at 10x weight, content at 1x, code at 0.5x
• Tokenizer: Porter stemmer + Unicode61 (handles diacritics)
• TF-IDF: scikit-learn TfidfVectorizer (term-weighting), unigrams + bigrams, code blocks stripped, top-10 keywords per conversation, min_df=2 for larger language groups and min_df=1 for small groups, max_df=0.8
• Language Detection: langdetect per message, 15 languages supported
• Parser: Canonical thread extraction via current_node backward traversal
• Code separation: Fenced code blocks extracted to separate field
• PUA cleanup: Unicode Private Use Area (PUA) citation markers stripped
• Citeturn cleanup: ChatGPT citation markup (citeturn0search1, etc.) stripped

Performance

Tested on 149MB export (1,514 conversations, 16,689 messages):

Anti-Patterns

Error Handling

Bundled Resources Index