transqlate 0.1.2

Natural Language to SQL CLI powered by a fine-tuned LLM.

Transqlate

Transqlate is a production‑ready natural language → SQL assistant powered by a fine‑tuned Phi‑4 Mini model. It lets anyone—technical or not—create and run complex queries over SQLite, PostgreSQL, MySQL, MSSQL, or Oracle databases using plain English.

---

Key Features

• Schema‑Aware NL→SQL – Automatically extracts your database schema and prunes it with an RAG pipeline so prompts fit within model limits.
• Interactive CLI – Generate queries, explore schemas, edit or write SQL, and dynamically switch connections with :changedb (alias :change_db).
• Safe Execution – DDL/DML statements require explicit confirmation. Lost connections trigger an optional reconnect with troubleshooting tips.
• Chain‑of‑Thought Reasoning – View the model’s reasoning and final SQL, converted from SQLite syntax to your database dialect with automatic schema qualification.
• Customizable Inference – Control maximum generation length (--max-new-tokens), enable Python tracebacks for debugging, and point to local or HF-hosted model weights.
• Offline‑Friendly – Runs on CPU and can pre‑download the sentence‑transformer embedding model so future runs work without internet access.
• Supports One‑Shot Queries – Use -q "question" for quick, non‑interactive SQL generation.

---

Installation

pip install transqlate

---

Usage

Launch the interactive CLI:

transqlate --interactive

Run a single question:

transqlate -q "List all customers who made purchases in March." --db-path path/to/database.db

REPL commands include :show schema, :history, :edit, :write, :examples, :clear, :changedb, :about, and :exit.

---

License

MIT License

---

Author

Shaurya Sethi Contact: [e-mail]

---

For the full user manual and source code, visit the GitHub repository or the Hugging Face model page.

---

What's New in v0.1.2

• Cross-dialect SQL post-processing: Automatically transforms model-generated (SQLite-style) SQL into the target dialect (PostgreSQL, MySQL, MSSQL, Oracle) with clear conversion notices.
• Enhanced CLI workflow: Added :edit for modifying generated SQL, :write for manual queries, and seamless database switching with :changedb/:change_db.
• Improved schema extraction: Robust support for PostgreSQL, Oracle, and MSSQL with qualified and escaped identifiers; Oracle system tables can now be filtered.
• Safer execution and state management: CLI now detects and recovers from lost connections, rolls back on errors, and requires confirmation for DDL/DML.
• History and troubleshooting: Query history now shows manual/edited status and execution results. Added :troubleshoot for step-by-step connection help.
• Token limit control: Set output length using the --max-new-tokens argument (default: 2048).
• Numerous bugfixes and UX refinements: Identifier quoting, error messages, startup feedback, and more.

For the full release notes, see CHANGELOG.md.