Thin Wrap

A terminal chat for any LLM API endpoint -- bridging pure coding and vibe-coding.

Installation

Linux/macOS (curl/wget + unzip required):

curl -fsSL https://raw.githubusercontent.com/thunderbyte-labs/thin-wrap/main/install.sh | sh

Installs to ~/.local/bin / ~/.local/lib/thin-wrap/, config at ~/.config/thin-wrap/, updates PATH.

Windows: Download .zip from Releases, extract, add to PATH manually.

Uninstall:

curl -fsSL https://raw.githubusercontent.com/thunderbyte-labs/thin-wrap/main/uninstall.sh | sh

Features

• Multi-LLM: Switch providers via /model (DeepSeek, Gemini, OpenRouter, custom endpoints).
• File Context: Ctrl+B opens a 3-column browser (editable / readable / navigator). r/e/d moves files; Ctrl+D clears all.
• Proxy: Use any SOCKS5/HTTP proxy (--proxy or /proxy). Per-model proxy hints via "proxy": true in config.
• Intelligent Editing: LLM-recommended edits get timestamped backups (file.thin-wrap.20250130.py) with Python-native diff stats. Backup can be disabled entirely via backup.enabled (see Configuration).
• Session Logging: Conversations auto-saved as .toml.zip with metadata (message count, preview). Reload with /reload.
• Free Chat: Choose "No root directory" at startup to chat without file context.
• History: PageUp/PageDown navigates sent messages and temporary drafts.

Configuration

config.json is resolved in order: --config arg → $THIN_WRAP_CONFIG_DIR → ~/.config/thin-wrap/ → executable directory.

Models

Each entry requires model, api_key (env var name), api_base_url.
Optional fields:

Field	Type	Default	Description
proxy	bool	false	Prompt for proxy when selecting this model
endpoint	string	/chat/completions	Custom API endpoint path
input_key	string	"messages"	Payload key for input (e.g., "input" for DashScope)
extra_arguments	object	{}	Additional JSON fields sent in request body
plugins	list or dict	[]	Define API plugins/tools (see Plugins section)

Example:

{
  "gemini-2.5-flash": {
    "model": "gemini-2.5-flash",
    "api_key": "GOOGLE_API_KEY",
    "api_base_url": "https://generativelanguage.googleapis.com/v1beta/openai/",
    "proxy": true
  }
}

Plugins

The plugins field lets you attach API-level plugins (e.g., web search, code execution) to a model. Plugins are merged into the request payload and are provider-dependent.

• List form: Simple list of plugin names or type objects. Currently only tested on OpenRouter.

  "plugins": ["web_search", "web_extractor"]

• Dict form: Full plugin definitions with options. Currently only tested on Qwen-Plus 3.6.

  "plugins": {
    "tools": [
      { "type": "web_search", "search_strategy": "agent" },
      { "type": "web_extractor" }
    ]
  }

Note: Plugins functionality depends on the API provider. For OpenAI-compatible endpoints, equivalent behaviour can be achieved via extra_arguments with a tools key. The plugins field provides a more abstract, model-centric way to declare them.

Backup

Controls file backup behaviour during intelligent editing.

When "enabled": true, the three fields timestamp_format, extra_string and overwrite_original are mandatory.

"backup": {
  "enabled": true,
  "timestamp_format": "%Y%m%d%H%M%S",
  "extra_string": "thin-wrap",
  "overwrite_original": true
}

Field	Type	Description
enabled	bool	Master switch. false disables all backups (direct overwrite).
overwrite_original	bool	If true: rename original to timestamped backup, then write new content to original path. If false: write new content to a separate timestamped file, leave original untouched.
timestamp_format	string	strftime format used in backup filenames.
extra_string	string	String inserted before the timestamp in backup filename.

Usage

1. Run thin-wrap (or python thin_wrap.py from source).
2. Select a project root (or free chat mode).
3. Choose a model (proxy prompt appears if needed).
4. Type message, press Alt+Enter to send.
5. Use Ctrl+B to manage files, PageUp/Down for history.

CLI Arguments

--root-dir <path>      Project root
--read <files>         Readable files (space-separated)
--edit <files>         Editable files
--message <text>       First message to send
--proxy <url>          Proxy URL
--config <path>        Config file path
--help                 Show help with data locations

Commands

Command	Description
/help /?`	Show command help
/clear	Clear conversation
/bye	Exit (saves session)
/model	Switch model (interactive or name)
/reload	Load a previous session
/rootdir	Change root directory (option 0 = free chat)
/files	Open file context menu
/proxy	Configure proxy (off to disable)

Architecture

Module	Role
thin_wrap.py	Entry point, chat loop
config.py	Config loading and validation
llm_client.py	Raw httpx-based API calls
file_processor.py	Query generation, XML parsing, file editing
input_handler.py	Multi-line input, history, command completion
command_handler.py	Slash command dispatch
menu.py	Textual-based file browser
proxy_wrapper.py	Proxy validation, SOCKS/HTTP
session_logger.py	TOML+ZIP session persistence
tags.py	XML tag utilities
text_utils.py	Text cleaning, token estimation
ui.py	Banners, colors, interactive selectors

Security

• Refuses root execution.
• API keys via environment variables (strongly recommended).
• XDG-compliant config storage.

Contributing

Maintain cross-platform compat, PEP 8, add tests for new features. PRs to main. Read TESTING.md.

License

AGPL-3.0