Nyayrithm docs
A graph of AI agents argues a case from evidence you supply. Each agent plays a
distinct legal role with its own knowledge scope and model, and every claim
resolves to the exact source passage behind it.
Where things live
| Surface |
What it is |
| This site |
Developer and self-hosting docs: setup, architecture, providers, deployment. |
| nyayrithm.aayushjoshi.dev |
The product landing page and in-app documentation. |
| nyayrithm.ai.aayushjoshi.dev |
The app itself: sign in, add evidence, run proceedings. |
What it does
| Capability |
Detail |
| 9 legal roles |
Judge, Prosecutor, Defense, Plaintiff, Accused, Witness, Investigator, Expert Witness, and a Custom role |
| Dynamic agent graph |
Agents spawn specialist sub-agents mid-proceeding; the orchestrator auto-spawns when a domain gap opens |
| Multi-modal evidence |
PDF, DOCX, audio (Whisper), video (ffmpeg and Whisper), images |
| Role-scoped retrieval |
Each agent retrieves only the evidence its role is permitted to see |
| Citations |
Inline [EVIDENCE:uuid:chunk_idx] markers resolve to the exact passage; each turn is marked cited, inferred, or disputed |
| Provider-agnostic |
Mix OpenAI, Anthropic, Gemini, and Ollama per agent, or run one free provider for the whole court |
| Any backend |
PostgreSQL, MongoDB, SQLite, or DynamoDB for data; Qdrant, Chroma, Pinecone, or pgvector for vectors. One env var each. |
Start here
The zero-cost path
A free Google Gemini key plus SQLite, Chroma, and sentence-transformers (or a
local Ollama model) is a complete stack with no external calls. See
Running locally.