Skip to content
Next Robot

Our platform · Russia

All cases

NextRobot Work — Corporate AI on Local Models

A full corporate workspace with 34 AI agents on local LLMs.

NextRobot Work: desktop screenshot

// Results

AI agents on local models
34
financial calculators
11
VRAM for the full local model set
28.8 GB

// How it works

  • 34 AI agents in one corporate workspace
    34 AI agents in one corporate workspace
  • AI estimator drafts a priced local estimate
    AI estimator drafts a priced local estimate
  • Local models run on your own server
    Local models run on your own server
  • Built-in messenger for project teams
    Built-in messenger for project teams

What we built

NextRobot Work runs Qwen-class models and BGE-M3 embeddings on Ollama, vLLM or llama.cpp, with hybrid RAG over company documents and local web search. A manager agent splits a request into up to 8 subtasks for agents and people.

  • 34 agents: legal, finance, procurement, HR, engineering
  • Manager agent for multi-step requests
  • Hybrid RAG with a legal corpus
  • Local call transcription with whisper.cpp

// Stack

  • Next.js
  • Local LLM (Qwen, vLLM, Ollama)
  • Hybrid RAG
  • whisper.cpp
  • LiveKit
  • Capacitor
// Next case
// Your project

Let's build this for you

Describe the task. An engineer replies within 24 hours with a plan and an estimate.

  • Free AI audit
  • Fixed-price pilot
  • NDA on request