Nexus platform

About

Open-stack reference RAG skeleton: ingestion, Gemini embeddings stored in Postgres with pgvector, cosine retrieval, and citation-aware answers returned as structured JSON usable from the REST API and MCP.

Runtime models (env overrides)

Chat / JSON generation: gemini-2.5-flash
Vision OCR (PDF / image): gemini-2.0-flash
Embeddings: gemini-embedding-001 · 768-dim vector column

Values shown reflect your environment at request time. Defaults live in lib/ai.ts when env vars are unset.

Chat response envelope

Version nexus.chat.v1 objects include:

text plain language string
markdown rich Markdown (GFM, KaTeX) with optional [[S1]] cite markers
sources[] chunk ids, filenames, similarity, excerpts
hallucination_flag when retrieval gate fails

Legacy fields answer and citations mirror text and sources for older clients.

Application stack

Next.js 14 App Router, TypeScript, Tailwind CSS
Prisma ORM on PostgreSQL with pgvector (IVFFlat cosine index via init route)
Google AI Studio: Gemini JSON chat, multimodal OCR, REST embedContent
Client rendering: react-markdown, remark-gfm, remark-math, rehype-katex
Optional MCP JSON-RPC surface at /api/mcp

Open Query Open Ingest Home