chat2html

Share-conscious HTML exporter for AI coding-assistant conversations (Claude and Codex).

Why chat2html?

• Four formats, one tool — Auto-detects Claude Code JSONL, claude.ai exports, claude-chat-exporter Markdown, and OpenAI Codex CLI JSONL.
• Safer sharing defaults — Tool results are omitted and OAuth URLs (with state, code, token, callback paths) are redacted by default. Use --full only when you need the full picture for yourself.
• Self-contained output — One HTML file with light/dark themes, syntax highlighting, and collapsible thinking blocks and long pastes.
• Zero install — uvx chat2html session.jsonl.

Supported input formats (auto-detected)

1. claude.ai export (conversations.json / .jsonl)

   • Downloadable from Settings → Privacy → Export data (inside the ZIP).
   • Contains multiple conversations — pick them by listing, searching, or index.

2. Claude Code session (~/.claude/projects/<proj>/*.jsonl)

   • Line-based logs with type / uuid / sessionId.
   • Renders tool-use history (Bash / Read / Edit / Agent, etc.).
   • thinking blocks and long tool_result outputs are collapsed into <details>.

3. claude-chat-exporter.js Markdown (.md)

   • See https://github.com/agarwalvishal/claude-chat-exporter.
   • Uses ## Human (date): / ## Claude: headers.

4. OpenAI Codex CLI session (~/.codex/sessions/*.jsonl)

   • Line-based logs with top-level {timestamp, type, payload}.
   • Renders user / assistant text, function_call + function_call_output pairs (e.g. exec_command), and custom_tool_call (e.g. apply_patch).
   • Encrypted reasoning is omitted; visible reasoning.summary is rendered as a thinking block.

Quickstart

Run directly from PyPI with uv — no install required:

uvx chat2html session.jsonl

The examples below use chat2html as shorthand for the command above.

Usage

# Auto-detect format (Claude Code JSONL / Markdown)
chat2html session.jsonl
chat2html conversation.md
chat2html session.jsonl -o out.html

# claude.ai export: list conversations
chat2html conversations.json

# claude.ai export: search by title
chat2html conversations.json -s "API"

# claude.ai export: convert by index
chat2html conversations.json -i 0,3,7 -d out/

# claude.ai export: convert all
chat2html conversations.json --all -d out/

# Batch multiple files (Markdown / Claude Code JSONL)
chat2html a.md b.jsonl -d out/

# Directory: pick logs interactively from a TUI list
chat2html ~/.claude/projects/myproj -d out/

# Directory: convert everything underneath without the picker
chat2html ~/.codex/sessions/2026/04 --all -d out/

Directory mode (TUI picker)

Pass a directory and chat2html will walk it for .md / .jsonl files, drop anything that isn't a supported chat log, and show a multi-select list (↑↓ to move, Space / x to toggle, Enter to confirm, Esc / q to quit). Entries are sorted newest-first by mtime, with a one-line preview of the first user message (housekeeping slash-commands like /clear are skipped so they don't become the preview text). Combine with --all to skip the picker and convert every log under the directory.

Press Tab in the picker to cycle the column focus through three modes: balanced view, filename-focused (drop the preview, expand the name to its relative path under the walk root so proj-a/session.jsonl and proj-b/session.jsonl are distinguishable), and snippet-focused (drop the filename to read more of the first user message — the preview cache holds up to ~250 characters).

Two safety caps keep the picker manageable when pointed at a deep tree like ~/.claude or ~/.codex:

• --depth N — max recursion depth from the given root (0 = root only, default 5).
• --max-files N — cap the list at the N most-recent files (0 = no cap, default 200). When entries are dropped, a notice is printed to stderr.

Options

⚠️ What chat2html does NOT protect against

chat2html redacts OAuth URLs and omits tool results by default, but it is not a general-purpose secret scrubber. Before sharing any output, you should still review it for:

• API keys and tokens (e.g., sk-ant-..., ghp_...) that may appear in tool inputs or assistant text
• Personal file paths (/Users/yourname/..., C:\Users\...)
• Internal hostnames, repository names, or IP addresses
• PII in pasted content (emails, phone numbers, etc.)
• Long pastes are collapsed into <details> but still present in the HTML source — they're hidden visually, not removed.

If you need stronger guarantees, consider running a secret scanner (e.g., gitleaks, trufflehog) on the output before sharing.

Extra safety: scan the output before sharing

chat2html does not detect arbitrary secrets like API keys or tokens embedded in conversation content. If you're sharing with a wide audience, pipe the output through a dedicated secret scanner:

• gitleaks — gitleaks dir out.html -v
• trufflehog — trufflehog filesystem out.html

Both are open-source CLIs available via Homebrew and most package managers.

Development

uv sync --all-groups        # install runtime + dev dependencies
uv run pytest               # run the test suite
uv run ruff check .         # lint
uv run ruff format .        # auto-format
uv run pre-commit install   # install the git hook (one-time per clone)

pre-commit runs ruff (lint + format) and a few hygiene checks (trailing whitespace, EOF newline, TOML/YAML syntax, large-file guard) on every commit; see .pre-commit-config.yaml. CI runs ruff check, ruff format --check, and pytest on every push and PR (see .github/workflows/ci.yml).