Indian law, understood — entirely on your machine.
Vidhi is a legal language model fine-tuned on approximately 100,000 Supreme Court judgments, High Court orders, and Central Acts. Built for legal research, statutory interpretation, and Indian jurisprudence — across criminal, civil, constitutional, and commercial law.
$ ollama run hf.co/prawinin/vidhi
Section 138 NI Act — dishonour of cheque for insufficiency of funds. On the facts you state, Vidhi sets out the statutory ingredients in order:
1. A cheque drawn on an account for a legally enforceable debt · 2. Presentation within validity · 3. Return unpaid for insufficiency or amount exceeding arrangement · 4. Written demand within 30 days of return · 5. Non-payment within 15 days of notice.
Illustrative output. Always verify against the bare Act and current case law before reliance.
What it does
A research assistant for Indian jurisprudence.
Vidhi is tuned for the way Indian lawyers actually work — provisions, ingredients, limitation, procedure, and the reasoning of constitutional, criminal, civil, and commercial benches. A research and educational tool, not a substitute for counsel.
Statutory interpretation
Parse Central Acts provision-by-provision: ingredients, exceptions, provisos, and cross-references — from the Constitution to the IPC, CrPC, CPC, NI Act, and Companies Act.
Case-law reasoning
Summaries and ratio-oriented discussion grounded in Supreme Court and High Court doctrine — holdings, tests laid down, and doctrinal evolution across benches.
Criminal & civil procedure
Step-throughs of CrPC and CPC workflows: cognizance, charge, bail, trial stages, pleadings, interim relief, execution — with limitation awareness.
Commercial & regulatory
Working knowledge of the Companies Act, Consumer Protection Act, Transfer of Property Act, and GST legislation for first-pass commercial analysis.
Drafting assistance
Structure legal notices, complaints, written statements, and research memos — with issue framing, relief mapping, and checklist-style completeness.
Private by design
Every query stays on your hardware. Client facts never leave the machine — suitable for chambers, clinics, and classrooms with confidentiality needs.
Training corpus
Trained on a large corpus of Indian legal documents.
Supreme Court judgments across constitutional, criminal, civil, and commercial benches · High Court orders and doctrine from multiple jurisdictions · Central Acts and statutory provisions covering major legislation in force.
Ratio decidendi, tests, and doctrinal lines — the core of Vidhi's reasoning patterns.
Procedural texture and divergent High Court readings that matter in practice.
Constitution, IPC, CrPC, CPC, NI Act, Consumer Protection, TPA, Companies Act, GST, and more.
Quick start
Running in under a minute.
Two supported paths. Ollama is recommended for most users; llama.cpp for direct control. No API keys, no accounts, no network calls at inference time.
Run with Ollama
Install Ollama, then pull and run. Loads in under 10 seconds on a 4 GB GPU, under 20 seconds on CPU.
ollama run hf.co/prawinin/vidhi
Model card
Small enough to run anywhere. Tuned for one job.
A LoRA fine-tune of Llama 3.2 3B Instruct, quantized for consumer hardware — built with Unsloth.
| Property | Value |
|---|---|
| Base model | Llama 3.2 3B Instruct |
| Fine-tuning method | LoRA · rank 16 · alpha 16 |
| Context window | 8,192 tokens |
| Quantization | Q4_K_M (GGUF) |
| File size | ~2.0 GB |
| Minimum hardware | 3 GB VRAM (GPU) or 8 GB RAM (CPU) |
| Languages | English · Hindi (legal usage) |
| Repository | huggingface.co/prawinin/vidhi |
Why 3B, and why local?
Large general models are powerful but impractical where confidentiality, cost, and connectivity matter. Vidhi trades breadth for depth: a compact model that carries Indian statute and precedent knowledge and runs wherever a lawyer, student, or researcher works.
- Chambers & courts without reliable internet
- Legal-aid clinics handling sensitive facts
- Classrooms teaching statutory reading
Responsible use
What Vidhi is — and isn't.
Vidhi is not a substitute for qualified legal advice from a licensed advocate. Do not rely on outputs without independent verification.
The model may not reflect recent amendments, new legislation, or judgments delivered after its training data was collected.
Always check outputs against bare Acts, official gazettes, and reported judgments before use in any professional context.
License & access
Personal, non-commercial evaluation.
Redistribution, sublicensing, sale, modification, or use in any product, service, or research publication requires explicit written permission.
Commercial licensing & partnerships
For enterprise deployment, commercial licensing, or collaboration, contact the author directly. Built with Unsloth; underlying base model subject to the Meta Llama 3.2 Community License.
| Permitted | Personal evaluation |
| Requires permission | Commercial use · redistribution · derivatives |
| Base license | Meta Llama 3.2 Community License |
FAQ
Questions, answered briefly.
How do I run Vidhi?
Install Ollama, then run ollama run hf.co/prawinin/vidhi. For direct control, use llama-cli -hf prawinin/vidhi with llama.cpp. Both run fully offline after download.
What hardware do I need?
The Q4_K_M build is ~2.0 GB. Minimum: 3 GB VRAM on GPU, or 8 GB RAM on CPU. A 4 GB GPU loads the model in under 10 seconds.
Which areas of law does it cover?
Criminal, civil, constitutional, and commercial law — including the Constitution, IPC, CrPC, CPC, NI Act, Consumer Protection Act, Transfer of Property Act, Companies Act, and GST legislation, plus Supreme Court and High Court doctrine.
Is it a substitute for a lawyer?
No. Vidhi is a research and educational tool. Outputs must be verified against authoritative sources and reviewed by a licensed advocate before reliance.
Can I use it commercially?
Not without permission. The model is released for personal, non-commercial evaluation only. Write to prawin@vyapai.tech for commercial licensing or enterprise deployment.
Does it send data anywhere?
No. Inference runs entirely on your hardware with no API, no internet connection, and no subscription. Your queries never leave the machine.