← Back to Discover

Hive

Open-source runtime where a persistent queen agent plans the work and hands it to parallel worker agents that record results in a shared SQLite tracker, with any model, MCP servers and skills.

Agent frameworksAutonomous agents
Visit website ↗

What it does

Hive is an open-source runtime for teams of AI agents that do repeatable business work. You describe an outcome to a long-lived queen agent that has a role and its own memory. She asks any clarifying question, does one unit of the job herself to learn what it takes, saves that method as a reusable skill, and then proposes a colony of parallel worker agents for you to approve. Each worker is a clone of the same agent loop with its own task, narrower tools and a strict budget. Results go into a shared SQLite tracker that the queen can query to see what is done and what is left, so progress lives on disk rather than in a context window. Playbooks run a proven method over every row with retries, backoff and a dead-letter list, and colonies can wake on cron, interval or webhook triggers. Agents save their state and resume after a crash or restart. Hive works with any model supported by LiteLLM, including OpenAI, Anthropic, Gemini, OpenRouter and local models through Ollama. Built-in tools cover shell commands, file editing, code search, PDFs, web scraping, charts and CSV files, and any MCP server can be added. It uses the open Agent Skills format and has a browser extension that lets agents work in your own Chrome. Setup needs Python 3.11+, Node.js 20+ and git, and a quickstart script opens a local dashboard. The site says that for a single prompt or one-off script a plain agent is simpler.

Capabilities

  • Queen agent with a role and markdown memory that spawns parallel worker agents
  • Shared SQLite tracker that survives restarts and can be queried or exported
  • Playbooks with retries, backoff and a dead-letter list for failing rows
  • Works with any model LiteLLM supports, including local models through Ollama
  • MCP servers, built-in tools and Agent Skills for extending what agents can do
  • Cron, interval and webhook triggers so colonies can run while you are away

When to use it

  • You have many similar units of work, such as leads, tickets or repositories, to process in parallel the same way each time
  • Work runs for hours or on a schedule and must survive a restart
  • You need results you can query and audit, not just read in a chat
  • You want to keep a person in charge of the decisions that matter