wtf is synthforge?

SynthForge is a command line tool that helps you create or enrich datasets for business intelligence and point of sale style demos using large language models. You point it at an existing file, a CSV, an Excel workbook, a Parquet file, or a DuckDB database table, describe the new columns you want added, write one prompt describing what should go in them, and the tool returns an enriched file or table. Behind the scenes it handles the prompt construction, the schema for the output, batching multiple rows efficiently, caching results, retrying failures, tracking metadata, and validating the results, along with setting up Azure OpenAI as a provider. The project positions itself as a middle ground between three existing approaches. Compared to simpler fake data libraries, it adds the ability to generate semantically meaningful fields using an actual language model, along with file and table augmentation, caching, and validation. Compared to statistical modeling tools meant to closely mimic real production data, it focuses on speed for demos and test datasets rather than faithfully reproducing sensitive data. Compared to writing your own one off scripts against an LLM, it keeps every prompt and run auditable, while letting normal deterministic code handle things like IDs, foreign keys, dates, and financial calculations. SynthForge is still an early project, currently at version 0.1. By default it uses a fake provider that needs no API key at all, so you can try the whole workflow offline before connecting a real model. The best supported real provider is Azure OpenAI's GPT-5 Nano, with an optional web search feature for looking up current facts like live prices. Other supported providers include Gemini with Google Search grounding and a Codex CLI based provider for local use. To install it, you clone the repository and install it with pip, requiring Python 3.11 or newer, with DuckDB included as a dependency. From there you can run a no SQL file augmentation command directly on a CSV, or work through SQL statements against a DuckDB database for more advanced relational work. The README stresses that this tool generates synthetic and demo data only, and it does not provide privacy guarantees or compliance certification, so it should not be treated as an anonymization solution for real sensitive data. The full README is longer than what was shown.