Chat Export Viewer

A Python CLI with no Python package dependencies that converts DeepSeek, Claude (Anthropic), and ChatGPT (OpenAI) conversation exports into styled HTML, Markdown, or cleaned JSON — with a built-in interactive single-page viewer.

Quick start: see QUICKSTART.md Pre-built binary (no Python needed): see BINARY_USAGE.md

Features

Multi-provider support — handles DeepSeek, Claude, and ChatGPT export formats out of the box
Template-based auto-detection — identifies the provider from JSON templates; if no template matches, the tool asks for explicit provider confirmation
Three output formats — styled dark-theme HTML, plain Markdown, and normalized JSON
Interactive SPA viewer — browse all exported conversations in one page with search, filtering, sort, and per-message jump navigation
Thinking blocks — DeepSeek THINK fragments and Claude extended thinking blocks rendered as collapsible sections
Web search results — DeepSeek SEARCH fragments displayed inline with titles, URLs, and snippets
No Python package dependencies — pure Python 3.10+ stdlib; nothing to install
Interactive mode — run with no arguments to be guided through zip extraction, provider detection, and export step by step
Extensible — drop a new template into provider_templates/ to add support for additional providers

Download pre-built binary

No Python required. Grab the latest conv-tool binary for your platform from the Releases page:

Platform	File
Linux (x86_64)	`conv-tool-linux`
macOS	`conv-tool-macos`
Windows	`conv-tool-windows.exe`

See BINARY_USAGE.md for setup and usage instructions.

Requirements (source install)

Python 3.10 or newer (uses match-free type hints; no third-party packages)

Quick verification

Run the repository smoke test (all three providers + SPA generation):

./scripts/smoke_test.sh

Installation

git clone <repo-url>
cd conversation-export-workbench

No virtual environment or Python package installation needed.

Getting your export file

DeepSeek

Open chat.deepseek.com → Settings → Export Data
Download the .zip archive — it contains conversations.json

Claude (Anthropic)

Go to claude.ai → Settings → Account → Export Data
Download the .zip — it contains conversations.json (and optional projects.json, users.json, etc.)

Usage

Interactive mode (recommended for first use)

python3 format_conversations.py

Scans the current directory for .zip archives and conversations.json files, prompts before each action, and optionally regenerates the SPA viewer at the end.

CLI mode

python3 format_conversations.py [options]

Options

Option	Default	Description
`--input FILE`	`conversations.json`	Path to input JSON
`--output DIR`	`output/<provider>/`	Where to write output files
`--provider NAME`	auto-detected	Force provider: `deepseek` \| `claude` \| `chatgpt`
`--format FORMAT`	`html`	Output format: `html` \| `md` \| `json`
`--id ID`	all	Export only the conversation with this ID
`--list`	—	Print all conversations with IDs and exit
`--combined`	—	Write all conversations to a single file (html/md)
`--yes` / `-y`	prompt	Overwrite existing files without prompting

Common examples

# List available conversations
python3 format_conversations.py --list

# Export all as HTML (auto-detects provider)
python3 format_conversations.py --format html --yes

# Export a single conversation
python3 format_conversations.py --id <uuid>

# Export from a specific file to a specific directory
python3 format_conversations.py --input ~/Downloads/claude-export.json \
    --output ~/Documents/claude-html --format html --yes

# One combined HTML file for all conversations
python3 format_conversations.py --format html --combined --yes

# Force provider when auto-detection is ambiguous
python3 format_conversations.py --provider deepseek --format md

Output formats

HTML

Styled dark-theme pages — one file per conversation. Features:

Thinking blocks as collapsible <details> elements
Web search results rendered inline
Per-message timestamps (toggleable)
Clean code blocks with language labeling

Markdown

Plain .md files suitable for editors, note-taking apps, or further processing. Thinking blocks are quoted, search results listed as links.

JSON

Normalized, provider-agnostic JSON:

{
  "id": "...",
  "title": "...",
  "started_at": "2025-06-01T09:00:00Z",
  "updated_at": "2025-06-01T09:05:00Z",
  "messages": [
    {
      "role": "user",
      "timestamp": "...",
      "parts": [{ "type": "text", "content": "..." }]
    },
    {
      "role": "assistant",
      "timestamp": "...",
      "parts": [
        { "type": "thinking", "content": "..." },
        { "type": "text", "content": "..." }
      ]
    }
  ]
}

SPA viewer

Generate an interactive single-page viewer from your exported HTML files:

python3 generate_spa.py --output output/ --yes

Then serve it locally (the SPA uses fetch() so it requires a real HTTP server):

python3 serve_spa.py
# Open the printed URL (first free port from 8080 up to 80890, clamped at 65535)

Or with explicit settings:

python3 serve_spa.py --host 0.0.0.0 --start-port 8080 --end-port 80890

SPA config and CSS templates

The SPA's appearance is controlled by CSS template files in config/spa_output_templates/ and wired together via config/spa.toml:

[spa]
main_template   = "config/spa_output_templates/main_spa.css"
thread_template = "config/spa_output_templates/thread.css"

[providers.deepseek]
thread_template = "config/spa_output_templates/deepseek_thread.css"

[providers.claude]
thread_template = "config/spa_output_templates/claude_thread.css"

[providers.chatgpt]
thread_template = "config/spa_output_templates/chatgpt_thread.css"

To use a custom config:

python3 generate_spa.py --config path/to/custom.toml --output output/ --yes

Viewer features

Feature	Details
Provider filter	Settings menu lets you narrow to DeepSeek, Claude, ChatGPT, or All
Live search	Filters sidebar list and highlights matches inside the open conversation; full-text search across loaded conversations
Sort	Newest first, oldest first, A→Z, Z→A
Jump navigation	Sidebar panel lists every user turn and assistant heading; scrolls to it on click; active entry tracks scroll position via IntersectionObserver
Timestamp toggle	Show/hide per-message timestamps across the whole viewer
Conversation visibility	Hide individual conversations from the settings menu
Scroll memory	Returns to your scroll position when switching back to a conversation
Lazy loading	Conversations are fetched on demand and cached in memory
Collapsible thinking	`<thinking>` blocks converted to `<details>` on load
Provider colour theming	DeepSeek (blue/green), Claude (violet/teal), and ChatGPT (OpenAI green/warm sand) accents applied automatically

Settings menu — provider filter, sort, timestamp toggle, conversation visibility

Provider auto-detection

The tool loads detection templates from provider_templates/ at startup. Each template is a JSON file named <provider>.conversations-template.json and contains a _detection_signature block:

{
  "_detection_signature": {
    "root_type": "array",
    "item_must_contain": ["chat_messages", "uuid"],
    "item_must_not_contain": ["mapping"]
  }
}

Detection logic: the top-level JSON must be an array; the first item must contain all keys in item_must_contain and none in item_must_not_contain. The first matching template wins. If no template matches, the tool does not guess from formatter heuristics and asks you to choose --provider (or prompts interactively).

Adding a custom provider

Create provider_templates/myprovider.conversations-template.json with a _detection_signature and _template_meta.provider set to your provider name.
Add a formatter module at formatters/myprovider.py implementing PROVIDER, ID_FIELD, TITLE_FIELD, build_html_single(), conv_to_md(), and build_json_single().
Register the module in _FORMATTERS in format_conversations.py.

Project structure

conversation-export-workbench/
├── format_conversations.py   # Main CLI entry point
├── generate_spa.py            # SPA builder CLI
├── formatters/
│   ├── __init__.py
│   ├── chatgpt.py             # ChatGPT formatter + active-branch tree walk
│   ├── claude.py              # Claude formatter
│   ├── deepseek.py            # DeepSeek formatter
│   ├── shared.py              # Shared utils: dates, slugify, markdown→HTML, HTML template
│   └── spa.py                 # SPA generator (CSS loader + metadata scanner)
├── config/
│   ├── spa.toml               # SPA config: which CSS templates to use per provider
│   └── spa_output_templates/
│       ├── main_spa.css       # SPA chrome: header, sidebar, menus, search
│       ├── thread.css         # Thread content area base styles
│       ├── deepseek_thread.css# DeepSeek accent colour overrides
│       ├── claude_thread.css  # Claude accent colour overrides
│       └── chatgpt_thread.css # ChatGPT accent colour overrides
├── provider_templates/
│   ├── chatgpt.conversations-template.json
│   ├── claude.conversations-template.json
│   └── deepseek.conversations-template.json
├── sample_data/
│   ├── chatgpt-convo.json     # Sample ChatGPT export (2 conversations)
│   ├── claude-convo.json      # Sample Claude export (3 conversations)
│   ├── deepseek-convo.json    # Sample DeepSeek export (3 conversations)
│   └── sample_output/         # Pre-rendered HTML + SPA from sample data
│       ├── index.html
│       ├── chatgpt/
│       ├── claude/
│       └── deepseek/
├── scripts/
│   └── smoke_test.sh          # Fast regression check for exports + SPA build
├── output/                    # Your personal exports (git-ignored)
├── QUICKSTART.md
└── README.md

Data model

DeepSeek export (`conversations.json`)

[ conversation, ... ]
  conversation:
    id, title, inserted_at, updated_at
    mapping: { node_id → node }
      node: { id, parent, children[], message }
        message: { files[], model, inserted_at, fragments[] }
          fragment types:
            REQUEST   → { type, content }
            RESPONSE  → { type, content }
            THINK     → { type, content }
            SEARCH    → { type, results[{ url, title, snippet, cite_index }] }
            READ_LINK → { type, url }

The conversation tree is walked from "root" following the first child at each node.

Claude export (`conversations.json`)

[ conversation, ... ]
  conversation:
    uuid, name, created_at, updated_at
    chat_messages: [ message, ... ]
      message:
        uuid, sender (human|assistant), created_at, updated_at
        content: [ block, ... ]
          block types:
            { type: "text",        text: "..." }
            { type: "thinking",    thinking: "..." }
            { type: "tool_use",    id, name, input }
            { type: "tool_result", tool_use_id, content }

Privacy

Path	Git status
`conversations.json`	Ignored
`.zip` / `data*.zip`	Ignored
`output/`	Ignored
`sample_data/`	Tracked (sample data only, no personal content)
`provider_templates/`	Tracked
`formatters/`	Tracked

Never commit your own conversations.json or export zips. The .gitignore is configured to block them by default.

License

MIT

Name		Name	Last commit message	Last commit date
Latest commit History 20 Commits
.claude		.claude
.github/workflows		.github/workflows
config		config
formatters		formatters
provider_templates		provider_templates
readme_assets		readme_assets
sample_data		sample_data
scripts		scripts
tests		tests
.dev-report-cache.md		.dev-report-cache.md
.forked-work-modified		.forked-work-modified
.gitignore		.gitignore
.mcp.json		.mcp.json
AGENTS.md		AGENTS.md
BINARY_USAGE.md		BINARY_USAGE.md
CLAUDE.md		CLAUDE.md
QUICKSTART.md		QUICKSTART.md
README.md		README.md
cli_main.py		cli_main.py
conv-tool.spec		conv-tool.spec
format_conversations.py		format_conversations.py
generate_spa.py		generate_spa.py
serve_spa.py		serve_spa.py

Folders and files

Latest commit

History

Repository files navigation

Chat Export Viewer

Features

Download pre-built binary

Requirements (source install)

Quick verification

Installation

Getting your export file

DeepSeek

Claude (Anthropic)

Usage

Interactive mode (recommended for first use)

CLI mode

Options

Common examples

Output formats

HTML

Markdown

JSON

SPA viewer

SPA config and CSS templates

Viewer features

Provider auto-detection

Adding a custom provider

Project structure

Data model

DeepSeek export (conversations.json)

Claude export (conversations.json)

Privacy

License

About

Topics

Resources

Uh oh!

Stars

Watchers

Forks

Releases 2

Packages 0

Uh oh!

Contributors

Uh oh!

Languages

DeepSeek export (`conversations.json`)

Claude export (`conversations.json`)

Packages