Skip to content

Latest commit

 

History

History
58 lines (35 loc) · 1.42 KB

docs.md

File metadata and controls

58 lines (35 loc) · 1.42 KB

MultilingualCL

Documentation

This page summarizes how to run MultilingualCL and what to expect from it.

Running The REPL

From the project root:

python -m multilingualcl.repl

The REPL reads natural-language requests and tries to resolve them to shell commands.

Examples:

  • list files
  • show current directory
  • create directory

How Command Resolution Works

MultilingualCL uses a fallback pipeline:

  1. LLM-based intent extraction (if Ollama is enabled and reachable)
  2. Rule-based semantic intent extraction
  3. Static translation lookup in YAML resources

If no command can be resolved, it prints Command not recognized..

Safety And Preview

By default, the REPL:

  • classifies command risk
  • shows a preview before execution
  • asks confirmation for risky commands

These behaviors are configurable using environment variables from ../README.md.

Supported Resources

Core resource files are stored under ../resources/:

  • en/actions.md and en/resources.md for semantic matching
  • en/commands.yaml and fr/commandes.yaml for command catalogs
  • yaml/command_map.yaml for static command translations

Platform Compatibility

The generated commands target Linux tools (ls, grep, chmod, etc.). On Windows, run in WSL, Git Bash, or Linux containers for best results.

Building Documentation From Scratch

Follow the instructions in build.md.