browse_cases
ChatGPTList cases by metadata facets, without a question. Requires at least one facet filter; an unfiltered browse is rejected. legal_domain is constrained by the schema; the valid values are: 'Accident Claims & Insurance', 'Civil & Property Law', 'Constitutional & Writs', 'Contempt of Court', 'Corporate, Commercial & Financial', 'Criminal Law', 'Election Law', 'Family & Personal Law', 'Labor & Service', 'Taxation & Revenue Law', 'Testamentary & Succession', 'Unclassified'. case_category EXACT values: 'Administrative & General Writs', 'Bail Applications', 'Bail & Quashing', 'Civil Appeals & Revisions', 'Civil Contempt', 'Commercial & Taxation', 'Criminal Appeals & Revisions', 'Election Petitions', 'Habeas Corpus Writs', 'Land Revenue', 'Matrimonial Disputes', 'Motor Accident Compensation', 'Public Interest Litigation', 'Service Matters', 'Special Acts Crimes', 'Unclassified', 'Wills & Probate'. Outcome: metadata_label (exact: 'Allowed'/'Dismissed'/'Partly'/'Unlabeled'); disposal_nature is the court's raw free-text disposal and matches inconsistently. Also: court — the High Court NAME (e.g. 'Patna High Court', 'Delhi', 'Punjab and Haryana') or its code; names are resolved for you and an unrecognised court returns an ERROR listing the valid ones, never an empty result. High Courts only — not the Supreme Court. year (the year the judgment was delivered) narrows another filter and is not accepted on its own, e.g. legal_domain='Criminal Law' + year=2024. This tool does not match a topic or facts; search_cases and metadata_filter_search do.
case_brief
ChatGPTAssemble an instant, source-tagged CASE BRIEF for one case (by 'doc:case:<cnr>', 'doc:pdf:<sha>', or a bare CNR): identity + citation, outcome / final order, facts, issue, holding, ratio (legal principles), key factors, statutes the judgment cites, chronology, and counsel — each section tagged by source ('corpus' = from the judgment, 'derived' = computed). Built only from the judgment record, with no web research. It does not include cross-court lineage, property/registry history or analysis of external documents. The sections returned depend on the account's access.
case_citations
ChatGPTFor one Indian judgment (by CNR, or 'doc:case:<cnr>'), return its CITATION NETWORK from the corpus: how many later judgments RELIED ON vs DISTINGUISHED it (each judgment counted once) (a counts-based 'good-law' signal — NOT a legal determination), plus the precedents THIS case itself relied on (with CNRs where resolvable). Grounded public-record data, cited by CNR; factual, not legal advice. Supreme Court coverage is complete; High Court coverage is being added (an HC case may return no citation record yet).
case_strength
ChatGPTGrounded, qualitative case-strength assessment for a fact pattern, for review with an advocate: not legal advice and not a prediction of the outcome. Retrieves similar Indian judgments, reads their arguments/reasoning/outcomes, and reports where the position looks strong vs weak (strong/moderate/weak, with reasons), the arguments that have worked, what the other side typically argues, and the outcome distribution across the retrieved cases — every case-specific claim cited by CNR. It never returns a win percentage or probability, and does not supply an opposing side's arguments where the record does not capture them. OPTIONAL user_side — which side of the matter the person is on, so a judgment the OTHER side won is not offered to them as a strength. One of: criminal:defence — the accused, or someone asking for them criminal:prosecution — the victim or complainant civil:claimant — the party making a claim civil:defendant — the party a claim is made against The person's own answer is the reliable source (criminal: 'Is the case against you or someone close to you, or did you report it?'; civil: 'Are you making the claim, or is it being made against you?'). Omit it when the side is not sure: omitting is always safe and changes nothing, while a wrong value removes points that genuinely favour the person.
fetch
ChatGPTFetch a single document by the 'id' returned from 'search'. Returns {id,title,text,metadata}.
get_case
ChatGPTFetch a case by id: 'doc:case:<cnr>' (whole case, all documents), 'doc:pdf:<sha256>' (one document), or a bare CNR. Returns metadata and, per access tier, the judgment text, rhetorical roles, and named entities.
metadata_filter_search
ChatGPTSearch for cases matching your question, narrowed by metadata. court/year/judge narrow (year = the year the judgment was delivered; year or year_from/year_to, not both) the search; metadata_label (the outcome: 'Allowed'/'Dismissed'/'Partly'/'Unlabeled'), disposal_nature, legal_domain and case_category filter the results; every returned case satisfies them (a 'note' is returned if fewer than k match). legal_domain takes one of the listed values. metadata_label is the reliable outcome filter; disposal_nature is the court's raw free-text disposal. To list cases by metadata without a question, browse_cases does that.
search
ChatGPTSearch the legal corpus and return a list of results ({results:[{id,title,text}]}; each title carries the court, year and CNR). Pass an 'id' from a result to 'fetch' to retrieve the full document.
search_cases
ChatGPTSearch Indian Supreme Court and High Court judgments for cases that match the facts, keywords or topics of your question. Returns ranked cases (cnr, court, year, relevance score, citation snippet). Optional filters: court (a High Court name or code, e.g. 'Allahabad' or '9_13'); year = the year the judgment was DELIVERED, either year or an inclusive range year_from/year_to (not both).
search_then_fetch
ChatGPTFind cases matching the facts, keywords or topics of your question AND return the top matches as full case records in one call. Returns {cases:[...]}. Heavier than search_cases; use when you want the case contents immediately.
browse_cases
Claudecase_brief
Claudecase_citations
Claudecase_strength
Claudefetch
Claudeget_case
Claudemetadata_filter_search
Claudesearch
Claudesearch_cases
Claudesearch_then_fetch
Claude
