Project Overview

Two Repositories, One Integrated Workbench

OmniDev uses the Knowledge Engine as an embedded Go module rather than a separate platform microservice.

┌──────────────────────── OmniDev ────────────────────────┐
│ Next.js web UI                                          │
│ Go gateway: auth · users · chat · knowledge · agents    │
│                    │                                    │
│                    └── embedded knowledge-engine module │
└────────────────────────┬────────────────────────────────┘
                         │
       PostgreSQL + pgvector · Redis · MinIO · Tika · Docker

The repositories must be cloned side by side because apps/gateway/go.mod uses a local replace for knowledge-engine. Knowledge Engine also has its own standalone HTTP API and MCP server for other applications.

Project Roles

Component	OmniDev	Knowledge Engine
Primary role	Team AI knowledge & coding workbench	Embeddable or standalone RAG engine
Backend	One Go/Gin gateway	Go HTTP API and MCP server
Frontend	Next.js 15 / React 19	None
Data	PostgreSQL + pgvector, Redis	PostgreSQL + pgvector, optional Redis cache
Documents	MinIO attachments, local or MinIO knowledge storage	Configurable document storage and Tika parsing
AI	Multi-model chat, embeddings, background agent planner	OpenAI, Gemini, and Ollama embeddings
Execution	Sandboxed background agents	No agent runtime

Current Use Cases

OmniDev

• Multi-model chat — Streaming conversations, attachments, images, BYOK, and external MCP tools
• Grounded answers — Upload documents and inject pgvector retrieval results into chat
• Background tasks — Run user-defined agents with workspace tools and isolated code execution

Knowledge Engine

• RAG applications — Provide document context to LLM applications
• Semantic search — Search knowledge bases with pgvector
• MCP tools — Expose knowledge operations to MCP clients
• Document management — Parse, chunk, embed, and retrieve documents

Product Boundaries

OmniDev does not currently provide an online IDE, visual workflows, billing, one-click deployment, or a bundled monitoring stack. Agent-created pull requests are planned but not implemented. There is no Kafka, Elasticsearch, Temporal, or 16-microservice deployment.