# Aden Hive

> One agent loop that clones itself into a colony of workers

- Category: Agents & Automation / Multi-Agent Systems
- Language: Python
- Licence: Apache-2.0
- Repository: https://github.com/aden-hive/hive
- Website: https://docs.adenhq.com/
- Canonical page: https://ai-tldr.dev/tools/aden-hive/

## In short

Aden Hive is a model-agnostic runtime for what the project calls colonies of agents. A colony is a Queen — the persistent, client-facing lead — plus as many worker agents as the job needs. It sits in the Multi-Agent Systems category of the Agents & Automation tools and is open-source, written in Python and licensed Apache-2.0.

## Overview

Aden Hive is a model-agnostic runtime for what the project calls colonies of agents. A colony is a Queen — the persistent, client-facing lead — plus as many worker agents as the job needs. You describe an outcome in plain English; the Queen does one unit of the work herself to prove the path, factors that into a reusable skill and playbook, then fans out worker clones to run it at scale.

The design choice underneath is that there is only one execution primitive. The Queen is an agent loop, and every worker is a clone of it with the same tools and model but its own task — so there is no graph to compile and no orchestration boilerplate. Coordination happens through a shared tracker ledger and a persistent task plan rather than a data buffer, which is also what makes work resumable: the loop parks its state to disk and picks up where it left off.

Hive positions itself as the harness layer rather than the agent layer — the parts you need once agents leave the prototype stage. That means crash-safe park and resume, budget enforcement with throttles and automatic model degradation, real-time observability and audit trails, and out-of-band human escalation through Sentinel, an account-bound Slack or Telegram channel any agent in the colony can park on. Models come through LiteLLM-compatible providers including OpenAI, Anthropic, Gemini and local Ollama models, and external systems connect as MCP tools.

## What it does

- Colonies: a persistent Queen spawns worker clones on demand for parallel, long-running work — no DAG to wire
- Shared tracker ledger and a persistent task plan as the coordination substrate, validated by the Queen with SQL
- Crash-safe park and resume, so a long run survives a restart instead of starting over
- Cost enforcement with budgets, throttles and automatic model degradation, plus real-time metrics
- Sentinel human-in-the-loop: the loop parks, notifies a human over Slack or Telegram, and resumes on their reply
- Model-agnostic through LiteLLM (including local Ollama models) with MCP tools, general compute use and browser use

## Getting started

Hive uses a uv workspace layout and is deliberately not installed with pip — running `pip install -e .` at the repo root produces a placeholder package that will not work. Use the quickstart script. You need Python 3.11+ and an LLM provider; ripgrep is optional but recommended on Windows.

1. Clone and run quickstart

   This creates the framework and tools environments, sets up the encrypted credential store, configures a default model interactively and opens the Hive interface in your browser.

```bash
git clone https://github.com/aden-hive/hive.git
cd hive

./quickstart.sh        # macOS / Linux
.\quickstart.ps1       # Windows, PowerShell 5.1+
```

2. Reopen the dashboard later

   Run this from the project directory.

```bash
hive open
```

3. Build your first agent

   Type the agent you want in the home input box. The Queen asks clarifying questions and works out a solution with you, then you run it from the Run button or by asking the Queen to run it.

4. Start from a template instead

   Click "Try a sample agent" to browse templates — run one directly, or build your own on top of it.

## When to use it

- Run a business process end to end with agents that survive crashes and can be audited afterwards
- Fan one proven procedure out across hundreds of items in parallel without hand-writing an orchestration graph
- Put a hard budget and a human approval gate around agents that touch production systems
- Connect agents to CRM, support, messaging and internal APIs as MCP tools shared by the whole colony