Slaze is an experimental agent that can generate code, set up projects and execute tools using large language models. The entry point for running the agent from the console is run.py.

• Uses OpenAI or OpenRouter API for LLM interaction.
• Supports a collection of tools such as Bash commands, project setup, code generation and more.
• Prompts are stored in the prompts/ directory and can be selected or created when the app starts.
• Logs are saved under logs/.
• Interactive prompt creation: you can edit existing prompt files or create a new one on launch.
• New prompts may be sent to the LLM for analysis to generate an expanded task definition when an OPENROUTER_API_KEY is available.
• Tools include WriteCodeTool, ProjectSetupTool, BashTool, DockerEditTool and PictureGenerationTool, and OpenInterpreterTool for generating files and images inside a Docker container.
• Tool execution results are logged to logs/tool.log and logs/tool.txt for later review.

1. Install Python 3.12 (see .python-version).
2. Install dependencies:

pip install -r requirements.txt

3. Provide an API key via environment variables or a .env file in the project root:

OPENROUTER_API_KEY=...  # or OPENAI_API_KEY
OPENAI_BASE_URL=https://openrouter.ai/api/v1  # optional override for OpenRouter

Execute:

python run.py console

To use the interactive web interface, run:

python run.py web

By default the server starts on port 5002. The CLI attempts to detect your machine's LAN IP so you can open a browser on another device if needed. Open your browser to the printed address (e.g. http://192.168.x.x:5002/).

For a detailed walkthrough see docs/web_interface_guide.md.

The script loads environment variables with dotenv and launches the AgentDisplayConsole. You will be prompted to choose a prompt from prompts/ or create a new one. After choosing a task, the agent runs, using tools and displaying output directly in the terminal.

When creating a new prompt you can edit the text before saving. If the OPENROUTER_API_KEY environment variable is set, the prompt text is first sent to an LLM to produce a more detailed task description. A project directory under repo/<prompt> is then created and the final task is saved to logs/task.txt. Tool activity and agent messages are streamed to the console and also logged to logs/ for later reference.

This project uses Python's built-in logging module for application logging.

Configuration:

• Logging is configured centrally in config.py.
• Key settings include console logging (level controlled by the LOG_LEVEL_CONSOLE constant, typically "INFO") and rotating file logging to logs/app.log (level controlled by the LOG_LEVEL_FILE constant, typically "DEBUG"). Log rotation (based on size and backup count) is set up for logs/app.log.

Usage:

• To use logging in a module, obtain a logger instance:

  import logging
  logger = logging.getLogger(__name__)

• Example log messages:

  logger.debug("Detailed debug information.")
  logger.info("An informational message.")
  logger.warning("A warning occurred.")
  logger.error("An error occurred.")
  logger.critical("A critical error occurred.")

• To log exceptions:

  try:
      # ... some operation ...
      pass
  except Exception as e:
      logger.error(f"Operation failed: {e}", exc_info=True)
  # Or, if the message is sufficient and you want the stack trace:
  # logger.exception("Operation failed")

Specialized File Operation Logging:

• For auditing file creations, modifications, and deletions, the project uses a specialized logger in utils.file_logger.py.
• This system logs detailed information, including file content (for non-binary files) or metadata (for images/binary), to logs/file_creation_log.json.
• This JSON log is distinct from the main application logging in logs/app.log and provides a structured audit trail for file manipulations.

Viewing Logs:

• Logs are output to the console (typically INFO level and above by default).
• Detailed logs (typically DEBUG level and above by default) are stored in logs/app.log.
• The file_creation_log.json in the logs/ directory contains the audit trail for file operations.

• run.py – Console runner that starts the agent and sampling loop.
• agent.py – Core agent implementation communicating with the LLM and tools.
• prompts/ – Text files describing tasks that can be selected at startup.
• tools/ – Implementation of Bash, project setup, code generation and other tools.
• system_prompt/ – Base system prompt and code prompt helpers.
• utils/ – Helper modules for displays, logging and Docker interaction.
• logs/ – Log files and cached constants generated during runs.

Some tools expect Docker to be available (container name defaults to python-dev-container). Ensure the container is running if you plan to use tools that execute commands inside Docker.

The Bash tool automatically converts commands to PowerShell when running on Windows to keep things cross-platform.

All environment variables can be placed in a .env file. The main ones are OPENROUTER_API_KEY/OPENAI_API_KEY and optionally OPENAI_BASE_URL.

Build the image from the project root:

docker build -t slaze .

Run the container, supplying your API key:

docker run -e OPENROUTER_API_KEY=YOUR_KEY -p 5002:5002 slaze

You may also use OPENAI_API_KEY and optionally OPENAI_BASE_URL instead of OPENROUTER_API_KEY. The web interface will then be available at http://localhost:5002/.