hive

OpenHive is a zero-setup, model-agnostic runtime designed for "colonies" of AI agents. In this architecture, a lead "Queen" agent manages the process and spawns "worker" clones to handle parallel, long-running tasks. Unlike traditional agent frameworks that require compiling a workflow graph, Hive uses a single execution primitive where the Queen and workers share tools and models, coordinating via a shared tracker ledger and persistent task plan. Key capabilities include: - Colony Management: Dynamic spawning of worker clones for scalable execution. - Production Harness: Built-in crash-safe park/resume functionality, cost enforcement, and deep observability. - Human-in-the-Loop: The Sentinel system allows agents to escalate tasks to humans via Slack or Telegram, pausing state until a response is received. - Integration: Supports 100+ LLM providers via LiteLLM (including OpenAI, Anthropic, and Google Gemini) and connects to business systems via MCP (Model Context Protocol). - Tooling: Includes general compute use and browser use via a native extension. The framework is built in Python and is intended for teams moving AI agents from prototypes to production workloads where auditability, uptime, and reliability are critical.