---
license: cc-by-4.0
language:
  - en
pretty_name: Open India Law
size_categories:
  - 10M<n<100M
task_categories:
  - text-retrieval
  - question-answering
  - text-classification
tags:
  - legal
  - india
  - case-law
  - legislation
  - rag
configs:
  - config_name: judgments
    data_files: "in_*_judgments.parquet"
  - config_name: legislation
    data_files: "in_*_legislation.parquet"
---

# Open India Law

Indian primary law, extracted and structured: every judgment of the Supreme Court of India and all 25 High Courts, the decisions of 15 tribunals and regulators, and Central, State and Union Territory legislation down to the individual section.

| | Volume | Period |
|---|---:|---|
| Court judgments | 12,848,644 | 1950 to 2025 |
| Tribunal and regulator matters | 813,168 | 1985 to 2026 |
| Enactments | 22,265 | 1806 to 2026 |
| Sections of legislation | 1,098,577 | individually searchable |

District and trial court decisions are **not** included.

Per-court, per-forum and per-jurisdiction tables are in the [repository](https://github.com/Vaquill-AI/open-india-law/blob/main/COVERAGE.md).

## What this adds over the source PDFs

The courts publish PDFs, very often scanned with no text layer. This dataset is the result of extracting, segmenting and normalizing them:

- **Full extracted text** of every judgment held.
- **Section-aware chunks** with `section_type` and `section_priority`, segmented on the document's own structure rather than a fixed window.
- **Character and page provenance** on every chunk (`char_start`, `char_end`, `page_start`, `page_end`), so a passage can be located in the source document.
- **Metadata**: court, bench, bench strength, judges, petitioner, respondent, disposition, decision date, date of registration.
- **Legislation to section level**, with status, hierarchy and amendment counts.

## Loading

```python
from datasets import load_dataset

judgments = load_dataset("vaquill/open-india-law", "judgments", split="train")
acts      = load_dataset("vaquill/open-india-law", "legislation", split="train")
kerala    = load_dataset("vaquill/open-india-law", data_files="in_kerala_judgments.parquet")
```

Bulk source PDFs are on the [mirror](https://oss-data-in.vaquill.ai/index.html) rather than here, since the tribunal set alone is 564 GB.

## Sourcing

Government-only. Every record traces to a court, tribunal or government publisher: India Code for legislation, the tribunals' and regulators' own portals, and parliamentary publishers. No commercial law reporter and no third-party aggregator material is included.

Not every official Indian body publishes on a `.gov.in` domain. The Reserve Bank of India uses `rbi.org.in`, the Pension Fund Regulatory and Development Authority `pfrda.org.in`, the Delhi Real Estate Regulatory Authority `erera.co.in`. Those are the official portals of the bodies named.

## What has been removed, and why

**Identifying detail that a statute bars from publication.** Indian courts anonymize sexual-offence victims in the large majority of cases, so this dataset does **not** exclude by statute category: doing so would remove roughly 180,000 judgments that are lawful to publish precisely because the court controlled the disclosure. What is redacted is the narrow band where identity actually leaks:

- names of relatives of victims and of protected children, where the court named them
- telephone numbers, email addresses
- Permanent Account Number, Indian Financial System Code and Aadhaar numbers, each only where the surrounding text identifies the number as such

That last condition matters. Those patterns are shape-only and Indian case numbers share the shapes: `WPCT0123456` is an Indian Financial System Code by shape and `ABCDE1234F` is a Permanent Account Number by shape. Masking on shape alone corrupts the judgment text, so a context word is required.

**Documents a court directed not be published**, including in-camera matters. Those sit outside the statutory exception this dataset relies on.

**Documents not sourced from the court.** A small number carried a third-party site's print-view furniture in the body, which means they came from that site rather than from the court. Note that a judgment *citing* a commercial reporter is ordinary and is kept: Indian courts cite Manupatra, SCC OnLine and AIR constantly, and those are the judge's own words.

Counts of what each rule removed are published with each snapshot.

## Limitations

- A snapshot is a **point-in-time archive**, not the current state of the law. Verify against the official source before relying on any provision or judgment.
- Tribunal material is indexed by case, not yet by full text. Most tribunal *documents* are procedural order sheets rather than substantive decisions: about 348,516 of 2,112,201 are reasoned decisions.
- Coverage is uneven by court and by year. The per-court tables show the real depth.
- This is **not legal advice**.

## Licensing

Indian statutes and judgments are Government works under section 17(d) of the Copyright Act 1957. Reproduction is lawful under the section 52(1)(q) exception rather than because the text is public domain, so **we hold no rights in the underlying text and grant none** — your right to use it comes from the statute.

What is licensed is the compilation: the extraction, structuring, normalization, identifiers and metadata, under **CC BY 4.0**.

> Structured Indian primary-law data from the Open India Law corpus by Vaquill AI
> (https://github.com/Vaquill-AI/open-india-law), used under CC BY 4.0.

Attribution is also a condition our sources impose: the eCourts policy and those of the National Company Law Tribunal and the National Company Law Appellate Tribunal all require the source to be prominently acknowledged. That is why this is CC BY rather than CC0.

Full terms: [data/LICENSE.md](https://github.com/Vaquill-AI/open-india-law/blob/main/data/LICENSE.md).

## Reporting a problem

If this dataset contains material that should not be public, write to **contact@vaquill.ai** with the subject line `Open India Law - redaction request`. Directions of Indian courts and tribunals are honoured. Because a published snapshot is a fixed artifact with published checksums, corrections are made by publishing a superseding snapshot rather than editing one in place.

## Reproducing it

Every scraper and the whole release pipeline are at [github.com/Vaquill-AI/open-india-law](https://github.com/Vaquill-AI/open-india-law) under Apache-2.0.