Back to projects

hivemind

Distributed AI worker system

Run autonomous AI workflows from the terminal.

View on GitHub

Demo

terminal
$ hivemind run "analyze diffusion model papers"
> spawn worker[1]
> spawn worker[2]
> spawn worker[3]
>
> worker[1] gathering sources
> worker[2] summarizing architecture
> worker[3] extracting math
>
> aggregating results...
>
synthesis complete

Problem

Single-agent systems hit limits quickly. Complex tasks need multiple agents, shared memory, and orchestration.

What it is

hivemind is a distributed AI worker system where:

  • tasks are decomposed
  • agents collaborate
  • memory is shared
  • results converge

Features

Distributed Workers

Multiple agents run in parallel.

Shared Memory

Agents read and write shared state.

Task Pipelines

Decompose and orchestrate workflows.

Tool Usage

Agents call tools and APIs.

CLI Control

Run from the terminal.

Architecture

User Task
Orchestrator
worker[1]
worker[2]
worker[3]
Shared Memory
Final Output

Example result

Task

analyze diffusion model papers

Synthesis (agents)

## Diffusion model papers — synthesis

**Architecture (worker[2])**  
Common pattern: noise schedule → UNet backbone → conditioning embedding.  
DDPM vs DDIM: discrete vs continuous steps.

**Math (worker[3])**  
Key equations: forward process q(x_t|x_0), reverse process p_θ(x_{t-1}|x_t).  
Score matching objective.

**Sources (worker[1])**  
Ho et al. (DDPM), Song et al. (DDIM), Rombach (LDM).

Why I Built This

I wanted multi-agent systems that behave like distributed computation instead of simple chat loops.

hivemind lets agents collaborate, spawn work, and converge on results.

Repository

View on GitHub

Links