pplyz 0.1.3

LLM-powered CSV data analyzer with structured output generation using LiteLLM (short for LLM Analyser)

pplyz

Source & issues: https://github.com/masaki39/pplyz

Minimal CSV→LLM→CSV transformer powered by LiteLLM and uv.

Requirements

• uv
  • macOS/Linux: brew install uv or curl -LsSf https://astral.sh/uv/install.sh | sh
  • Windows: scoop install uv
• At least one LiteLLM-compatible API key (OpenAI, Gemini, Anthropic, Groq, etc.)

uvx downloads the right Python runtime automatically, so no global Python is needed once uv is installed.

Quick run (uvx)

uvx pplyz \
  data/sample.csv \
  --input question,answer \
  --output 'score:int,notes:str'

• --preview dry-runs a handful of rows (set [pplyz].preview_rows to change how many rows are shown).
• --model provider/name overrides the LiteLLM model (e.g., groq/llama-3.1-8b-instant).
• Prompts are entered interactively at runtime (history is stored under ~/.config/pplyz/). For non-interactive runs, provide the prompt when the CLI asks for it.

pplyz overwrites the input CSV; copy it first if you need to keep the original file.

Run uvx pplyz --help for every flag.

Common options

Flag	Description	Required
INPUT (positional)	Input CSV path.	Yes
-i, --input title,abstract	Comma-separated source columns passed to the LLM.	Yes (unless [pplyz].default_input is set)
-o, --output 'score:int,notes:str'	Output column schema. Types: bool, int, float, str (missing :type defaults to str).	Yes (unless [pplyz].default_output is set)
-p, --preview	Process a few rows and show would-be output without writing (row count configured via [pplyz].preview_rows).	No
-m, --model provider/name	LiteLLM model (default gemini/gemini-2.5-flash-lite).	No
-f, --force	Disable resume mode; always recompute rows and overwrite existing output.	No

Configuration

1. Create the user config once:

   mkdir -p ~/.config/pplyz
   $EDITOR ~/.config/pplyz/config.toml
   

2. Add only the providers you actually use:

   [env]
   OPENAI_API_KEY = "sk-..."
   GROQ_API_KEY = "gsk-..."
   
   [pplyz]
   default_model = "gpt-4o-mini"
   default_input = "title,abstract"
   default_output = "relevant:bool,summary:str"
   

3. At runtime pplyz loads settings in this order: environment variables → config file. The default path is ~/.config/pplyz/config.toml (or %APPDATA%\\pplyz\\config.toml on Windows; if XDG_CONFIG_HOME is set, it uses that). To keep configs elsewhere, set PPLYZ_CONFIG_DIR=/path/to/dir and place config.toml there.

Tip: pplyz data/papers.csv --input title,abstract --output 'summary:str' uses the positional data/papers.csv as the CSV input.

Settings reference

[pplyz] table

key	description	default
default_model	Sets the fallback LiteLLM model when --model is omitted.	gemini/gemini-2.5-flash-lite
default_input	Comma-separated columns used when -i/--input is omitted.	unset
default_output	Output schema used when -o/--output is omitted.	unset
preview_rows	Number of rows used when --preview is set (can also be overridden via PPLYZ_PREVIEW_ROWS).	3

Provider API keys

Set these inside the [env] table of your config.toml:

Provider	Keys (checked in order)
Gemini	GEMINI_API_KEY
OpenAI	OPENAI_API_KEY
Anthropic / Claude	ANTHROPIC_API_KEY
Groq	GROQ_API_KEY
Mistral	MISTRAL_API_KEY
Cohere	COHERE_API_KEY
Replicate	REPLICATE_API_KEY
Hugging Face	HUGGINGFACE_API_KEY
Together AI	TOGETHERAI_API_KEY, TOGETHER_AI_TOKEN
Perplexity	PERPLEXITY_API_KEY
DeepSeek	DEEPSEEK_API_KEY
xAI	XAI_API_KEY
Azure OpenAI	AZURE_OPENAI_API_KEY, AZURE_API_KEY
AWS (Bedrock/SageMaker)	AWS_ACCESS_KEY_ID, AWS_SECRET_ACCESS_KEY
Vertex AI	GOOGLE_APPLICATION_CREDENTIALS

Supported models

For the latest list of supported models, see the LiteLLM provider docs: https://docs.litellm.ai/docs/providers

Examples

Sentiment pass with a preview first (preview_rows set to 5 in your config):

[pplyz]
preview_rows = 5

uvx pplyz \
  data/reviews.csv \
  --input review_text \
  --output 'sentiment:str,confidence:float' \
  --preview

Boolean classifier that writes back into the same CSV:

uvx pplyz \
  data/articles.csv \
  --input title,abstract \
  --output 'is_relevant:bool,summary:str'

Model override with Anthropic:

uvx pplyz \
  data/papers.csv \
  --input title,abstract \
  --output 'findings:str' \
  --model claude-3-5-sonnet-20241022

Tips

• Boolean output columns keep binary classifiers deterministic (true/false).
• Some models do not support JSON mode; pplyz only sends response_format to models that advertise support. Explicitly state “return valid JSON only” in your prompt to keep outputs consistent.
• Keep prompts short and explicit about the JSON schema you expect to avoid parsing errors.
• Use --preview before long or expensive CSV batches to validate prompts and model choice.
• Resume mode is on by default; rows with existing output columns are skipped. Use --force to recompute everything.
• Dynamic (schema-less) mode is not supported; always provide --output (or set [pplyz].default_output).
• CSV encoding is UTF-8 only; convert input files beforehand if they use another encoding.