Open-source Indian legal information assistant for citizens. Dharmiq explains rights and obligations in plain language, grounded in statutory documents (IndiaCode corpus), with citations and clear disclaimers that it does not provide legal advice.
Get running in minutes → docs/setup.md
Generic chatbots are excellent generalists. They are a poor substitute for current, citable Indian law.
A model’s training cut-off is not a living statute book. It will not reliably have the latest India Code instruments, amendments, repeals, or case-law coverage — and it has no obligation to show you the provision it is paraphrasing. When the law changes (IPC → BNS, amended acts, supersessions), fluent confidence is often worse than silence.
Dharmiq is built as a grounded legal information system, not a chatbot with legal vibes:
- Living corpus, not stale memory — Answers retrieve from an IndiaCode corpus (and related eval / quality datasets) that we keep current; ingestion runs on a daily cadence so citations track what the statute book says now, not what a model remembered months ago.
- Citations are first-class — Rights, obligations, and penalties must trace to a source the user can open. Fluent but unverified claims are not the product.
- Async thinking, not a one-shot reply — Chat runs as a background agent pipeline (clarify → retrieve → answer → cite → validate) with live progress. Spending seconds or minutes to rethink, rerank, and verify is preferred over a fast guess under a tight sync latency budget.
- Clarify before guessing — Underspecified questions get structured follow-ups instead of confident wrong paths.
- Refuse over invent — Weak retrieval or unsupported statutory claims block release; honesty about ambiguity and conflicting sources beats false certainty.
Those tradeoffs — trust and currency over speed, breadth, and cost — are the
point. The fuller statement of principles, anti-goals, and tie-breakers lives in
docs/principles.md.
- Citizens who want plain-language explanations of rights and obligations, with sources they can check
- Legal Firms who want to run a grounded Indian legal RAG stack without trusting a closed SaaS for rights information
Dharmiq is early. The corpus is centered on allowlisted central statutes (IndiaCode), not the full judicial universe. Case-law / judgment coverage is limited compared with statutes. It is legal information, not a lawyer — when the stakes are real, consult a qualified advocate.
| Doc | Contents |
|---|---|
docs/setup.md |
Prerequisites, quick start (host / Docker), config, data ingestion |
docs/architecture.md |
Tech stack, repo layout, Mermaid system / agent diagrams |
docs/ops.md |
Monitoring (Prometheus/Grafana/Tempo, Temporal UI) and evaluation harness |
docs/CHANGELOG.md |
Version history |
docs/deployment.md |
Production deployment (Ubuntu + Nginx, Docker stacks) |
docs/principles.md |
Design principles and anti-goals |
backend/README.md |
API, agents, ingestion, eval details |
frontend/README.md |
Vite / Ashoka UI |
- Follow
docs/setup.md(host or Docker) - Read
docs/principles.mdbefore inventing product tradeoffs - Open a PR against Ankk98/dharmiq — issues and discussion welcome
Dharmiq provides general legal information, not legal advice. Users should consult a qualified lawyer for decisions that matter to them.