Documentation

Oracus.ai is a self-hosted Q&A surface for your product. Point it at your git repositories — Oracus.ai auto-extracts specs and dependency maps — and layer on Jira, JUnit, Confluence, OpenAPI, GitHub, and uploaded files. Ask anything in natural language; get answers that cite the source they came from.

Two flagship flows ship out of the box as opinionated examples — both are just the general Q&A engine with a tighter prompt:

1. Bug coverage: given a bug ticket, is the scenario covered by existing tests?
2. Spec ↔ implementation Q&A: natural-language queries that distinguish what the spec says from what the tests verify.

Get started

• Self-host Oracus.ai (request access)
• Configure AI providers and tasks

Architecture in one paragraph

Node 22 + TypeScript. Fastify for the API, with Zod schemas shared between server and frontend (no codegen, no drift). Postgres 16 with pgvector for embeddings. Prisma 6. pg-boss for the embedding job queue (no Redis). React 19 + Vite + Tailwind v4 for the app. tsyringe for class-based DI everywhere. The Vercel AI SDK abstracts the model providers so you can swap OpenAI, Anthropic, Google, or any OpenAI-compatible endpoint per task.