Skip to content

Command Line Interface (CLI)

🔧 Installation

# From the workspace root
cargo install --path panini-cli

# Or via cargo run (development)
cargo run -p panini-cli -- [commands]

Success Output: 

Finished dev [unoptimized + debuginfo] target(s) in 0.01s
Running `target/debug/panini [commands]`

🚀 Configuration (panini.toml)

The CLI uses a TOML configuration file to store your API keys and model preferences.

# panini.toml
provider     = "google"           # openai | anthropic | google
model        = "gemini-2.0-flash"
language     = "pol"              # pol | tur | ara
api_key      = "$GEMINI_API_KEY"  # Environment variable support
prompts_file = "panini-cli/prompts/default.yml"

🏗 Core Commands

1. extract

The main extraction tool.

  --target studentka --target czyta

Extraction Result (Simplified): 

{
  "word": "studentka",
  "morphology": { "lemma": "studentka", "pos": "noun", "case": "nominative" }
}

Useful Options: - --components: Comma-separated list (e.g., morphology,leipzig). - --temperature: Adjust LLM creativity (defaults to 0.2). - --ui-language: Language for pedagogical explanations (Defaults to English).

2. languages

Displays the list of all ISO codes supported by your current installation.

panini languages
# => Supported languages: pol, tur, ara, fra...

3. add-language

Launches the LLM-assisted process for generating the code for a new language.

  --config panini.toml

Generation Output: 

[INFO] Analyzing Japanese morphology...
[INFO] Generated src/japanese.rs
[INFO] Registered 'jpn' in registry.rs

LLM Scaffolding Warning

The add-language tool is a powerful automation that relies on an LLM to describe linguistic types. - Verify Everything: You must always manually verify the generated Rust types and extraction directives. - Use High-End Models: Always use "Pro" or "Ultra" class models (e.g., Gemini 1.5 Pro, GPT-4o, Claude 3.5 Sonnet) for the best results. - Linguistic Review: We strongly recommend having the generated configuration reviewed by a professional linguist or native speaker.

🪄 Productivity Tips

Pipes and JSON Output

The panini binary outputs pure JSON to stdout. You can use jq to filter results:

panini extract --text "..." --target "..." \
  | jq '.morphology.target_features'

Logging

Use the RUST_LOG environment variable to see prompt assembly details or API calls:

RUST_LOG=info panini extract --text "..."

Scaffolding with the CLI

The add-language command doesn't just generate an .rs file; it also attempts to register the language in registry.rs. Always check the changes after execution.