Hive

A self-hosted AI agent runtime coordinates parallel workers, tracks results locally, and supports Ollama or cloud models.

Hive coordinates AI agents for teams processing batches of leads, tickets, repositories or documents. A persistent lead agent, called a queen, develops a method on one item before assigning parallel workers, with human approval before the group starts.

Parallel work: Workers follow a reusable method while the queen stays available in chat and reviews their results.
Auditable results: A shared tracker lets you query, export and check completed work, then resume unfinished items after an interruption.
Memory and skills: Agents retain notes across sessions, recall relevant conversations and save proven procedures as reusable skills.
Scheduled tasks: Jobs can start on a schedule or webhook, recover after a restart and request human input through the Hive inbox, Telegram or Slack.
Tools and integrations: Agents can work with files, PDFs, web pages and your logged-in Chrome browser; MCP connects services such as GitHub, Gmail and HubSpot.

Hive is open source under Apache License 2.0 and runs locally on macOS, Linux and Windows, with a browser dashboard. Sessions, trackers and memory stay on your machine. It supports local models through Ollama without an API key, plus OpenAI-compatible endpoints and providers supported by LiteLLM, including Anthropic, OpenAI and Gemini. Choosing a cloud model sends model requests to that provider; image generation uses Hive LLM.