蜂群 fēngqún — a swarm of bees

Many small minds.One safe hive.

FENGQUN turns the HarmonyOS devices around you into a swarm of small on-device AI agents. Each one has a role. Together they form one private, resilient, post-quantum encrypted system — and nothing ever leaves the hive.

See the hive workSend a signal
  • Intelligent Experiences
  • Human-Centric: privacy by default
Nodes in hive● 5 online
Cloud calls0
ChannelML-KEM
Signals sent0
Move to guide · click to signal

01The problem

A lone bee doesn't survive the winter.

Your AI leaves home. Your hardware sits idle. Today's assistants work alone — and far away from you.

Leaving home

Cloud-first assistants

Questions, notes and voice recordings are shipped to someone else's servers. Even a shopping list costs you your privacy.

Sitting idle

Idle compute everywhere

A typical home owns a phone, tablet, laptop, TV and watch — a mesh of capability that no mobile OS lets you combine into one local AI resource.

Stored for later

Harvest now, decrypt later

Traffic captured today can be stored and decrypted by future quantum computers. Even home traffic deserves post-quantum protection.

02How the hive works

Every agent has a role.

Honest scope: N devices run N models in parallel with task routing — we don't "glue RAM", we schedule. One request, the best device answers.

YOUSCOUTSDANCERSFORAGERSWORKERSGUARDS

01 / 05Scouts · discovery

  1. 01Scouts

    DiscoveryAutomatic mesh of HarmonyOS devices nearby

    Scouts find each other in the local network — phones, tablets, laptops, TVs, watches — and every new device simply joins the swarm.

  2. 02Dancers

    Resource reportingLive RAM / compute / battery reporting per node

    Like the waggle dance, every node advertises its idle capacity: free RAM, compute, battery level.

  3. 03Foragers

    Task routingOne request, the best device answers

    Each AI request is routed to the device with the most available power right now.

  4. 04Workers

    On-device inferenceSLM / LLM / STT inference runs locally

    Small LLMs / SLMs and speech-to-text run fully locally. Zero cloud calls.

  5. 05Guards

    PQC channelML-KEM encrypted inter-device channel

    All node-to-node traffic is encrypted with post-quantum cryptography (ML-KEM).

03Demo use cases

Send a signal. Watch the swarm answer.

Narrow, real, end-to-end. Pick a request — the swarm finds the strongest device, answers on-device and keeps every byte at home.

Live simulationIdle

Choose a request

    Sample output

      Use case 01

      Private shopping list

      Speak or type "plan dinner for four and give me a shopping list" — a local model structures the list. The request never leaves your network.

      Use case 02

      Local notes search

      "What did I want to discuss with Paul?" — FENGQUN searches your notes on-device and drafts the answer from your own context.

      04Security & privacy

      Guards at every entrance. Nothing to leak.

      • Data stays home

        No cloud dependency at all. Works fully offline.

      • Post-quantum channel

        ML-KEM (Kyber) between every pair of nodes.

      • Minimal footprint

        No secrets in repo, minimal permissions, least surface.

      SimulationHive wall · sealed
      5 nodes · 10 ML-KEM pairsDeflected · 0

      05Platform fit · Why HarmonyOS?

      The hive was already built.

      FENGQUN could not be ported to another OS "without changes" — device clustering and distributed scheduling are platform-native here.

      Distributed capabilities

      The platform natively understands device clusters (SoftBus / distributed services).

      Whole-ecosystem devices

      Phones, tablets, laptops, TVs, watches, smart-home.

      On-device AI

      As a first-class platform concern.

      European angle

      Oniro & digital sovereignty: auditable, patchable, vendor-neutral.

      • ArkTS + ArkUI
      • API ≥ 20 (target: 6.1.1 / API 24)
      • HarmonyOS / OpenHarmony
      • On-device runtime (MindSpore Lite / llm.cpp — in evaluation)
      • ML-KEM (Kyber) via C/C++ NDK
      • DevEco Studio · hvigor · HDC

      06Fair questions

      The ones the jury will ask.

      Zero configuration — you already own these devices. Your private data stays on personal devices instead of a server in the basement. And it follows the natural topology of the HarmonyOS ecosystem you live in.

      Narrow tasks done well beat everything done poorly. Routing picks the strongest device per task — a laptop can serve a bigger model when it's idle.

      Offline is a bonus, not the selling point. Privacy is the selling point: nothing you ask ever leaves your home.

      07The team

      Five bees, one swarm.

      • Bee 01Kacper Kleczaj

      • Bee 02Filip Pecyna

      • Bee 03Mateusz Jędrzejczak

      • Bee 04Piotr Gliszczyński

      • Bee 05Janusz Marszałek

      蜂群HackYeah 2026 · Huawei Challenge: Imagine What's Next

      Small minds. Safe hive.

      Send a signalBack to the top
      %%SCRIPT%%