"# DevAssist"

DevAssist

A lightweight command-line AI assistant for programming help — code review, debugging, and general language/library questions — built on top of OpenRouter, with automatic fallback across multiple free models so you're not stuck when one is overloaded or rate-limited.

Features

• Multi-model fallback — configure a list of models; if one fails or is overloaded, DevAssist automatically retries with the next one
• Streaming responses rendered with proper Markdown formatting (headers, tables, syntax-highlighted code blocks) instead of raw text
• Repetition-loop protection — detects and stops runaway/degenerate model output automatically
• File-aware conversations
  • @filename — inline-include a file's contents in your message
  • /file <path> [question] — send a full file plus an optional question in one line
• Automatic code file generation — ask for a script, and DevAssist can save the generated code straight to a directory of your choice with /savecode
• Clipboard integration — /copy grabs the last response, /paste sends your clipboard content as the next message
• Persistent conversation history with configurable context length, saved to disk between sessions
• Slash commands for switching models, clearing/viewing history, and exporting chats to Markdown
• Rich terminal UI — spinners, colored output, tables, and syntax-highlighted code

Requirements

• Python 3.10+
• An OpenRouter API key

Installation

1. Clone or download this repository.
2. Install dependencies:

   pip install -r requirements.txt
   

3. Create a .env file in the project folder:

   OPENROUTER_API_KEY=your_key_here
   

4. Run it:

   python ask.py
   

Optional: run it from anywhere

Add the project folder to your system PATH and create a small wrapper script (ask.bat on Windows / a shell script on macOS/Linux) so you can just type ask from any terminal, in any folder.

Configuration

Settings live in config.json (auto-created on first run) and can also be changed live with slash commands:

{
    "default_model": "nvidia/nemotron-3-ultra-550b-a55b:free",
    "models": [
        "nvidia/nemotron-3-ultra-550b-a55b:free",
        "poolside/laguna-m.1:free",
        "poolside/laguna-xs-2.1:free",
        "cohere/north-mini-code:free"
    ],
    "temperature": 0.2,
    "history_limit": 10,
    "stream": true,
    "reasoning": true,
    "timeout": 120,
    "output_dir": "./generated"
}

Commands

Command	Description
/help	Show all commands
/clear	Clear conversation history
/history	View conversation history
/models	List configured models
/model <n>	Switch the active model
/save	Export the chat to Markdown
/copy	Copy the last response to the clipboard
/paste	Send clipboard content as your next message
/file <path> [question]	Send a full file, optionally with a question
/outdir <path>	View or set the folder where generated code is saved
/savecode [name]	Save code block(s) from the last response to disk
/quit	Exit DevAssist

You can also include any file inline in a message with @filename.

Project Structure

.
├── ask.py                 # Entry point / main loop
├── config.py               # Config loading, saving, and defaults
├── history.py               # Conversation history persistence
├── ui.py                     # Terminal rendering (Rich-based)
├── commands.py                 # Slash command handling, file loading
├── openrouter_client.py           # OpenRouter API client with fallback/retry
├── requirements.txt
└── config.json                        # Generated on first run

How It Works

DevAssist sends your conversation history to OpenRouter, streaming the response back token by token. If a model fails, times out, or returns nothing, it's automatically retried and then swapped for the next model in your configured list — so a single overloaded free-tier model doesn't block you from getting an answer.

When a response includes a full file (marked internally with a File: path/to/file.ext label), /savecode will write it straight to your configured output directory, creating folders as needed.

License

MIT — do whatever you'd like with this.