Database-driven agent graphs
Define AI workflows as rows, not code. Swap processing pipelines per organization without redeploying anything.
Arcnem Vision
Upload a photo, and a swarm of AI agents — orchestrated by LangGraph, connected through MCP, and configured from a database — will generate embeddings, write descriptions, and make everything searchable by meaning.
Features
Section titled “Features”GenUI chat interface
The AI generates real Flutter widgets at runtime — cards, galleries, interactive components — composed from JSON.
On-device Gemma
Intent parsing happens locally on the phone before anything hits the network. Private by default.
CLIP vector search
Images and descriptions embedded in the same 768-dimensional space. Search by image, by text, or by vibes.
Visual workflow builder
A React dashboard where you drag-and-drop agent graphs: workers, tools, supervisors, edges, the whole thing.
MCP tools as a first-class primitive
Five registered tools following the open Model Context Protocol standard. Agents call them. You can too.
Tech Stack
Section titled “Tech Stack”| Layer | Tech | What it does |
|---|---|---|
| Client | Flutter, Dart, flutter_gemma, GenUI, fpdart | Camera capture, on-device LLM, AI-generated UI, functional error handling |
| API | Bun, Hono, better-auth, Inngest, Pino | REST routes, presigned uploads, durable job scheduling, structured logging |
| Dashboard | React 19, TanStack Router, Tailwind, shadcn/ui | Workflow builder, document viewer, admin interface |
| Agents | Go, Gin, LangGraph, LangChain, inngestgo | Graph-based agent orchestration, ReAct workers, step-level tracing |
| MCP | Go, MCP go-sdk, replicate-go, GORM | CLIP embeddings, description generation, similarity search tools |
| Storage | Postgres 18 + pgvector, S3-compatible, Redis | Vector indexes, object storage, session cache |