OpenHive (Hive)

OpenHive is a zero-setup, model-agnostic execution harness that dynamically generates multi-agent topologies to tackle complex, long-running business workflows without requiring orchestration boilerplate. By simply defining your objective in natural language, the runtime compiles a strict, graph-based execution DAG that safely coordinates specialized agents to execute concurrent tasks in parallel. Backed by persistent, role-based memory that intelligently evolves with your project's context, OpenHive ensures deterministic fault tolerance, deep state observability, and seamless asynchronous execution across whichever underlying LLMs you choose to plug in. It is fully open-source under the Apache License 2.0 and backed by Y Combinator.

• Multi-Agent Coordination — parallel task execution across specialized agents coordinated via a graph-based DAG
• Auto-Generated Agent Graphs — describe your goal in plain English and the system automatically generates the agent topology, connection code, and test cases
• Role-Based Memory — persistent memory that evolves with your project's context for deterministic, stateful execution
• Zero Setup — no technical configuration required; run ./quickstart.sh to get started on macOS/Linux or .\quickstart.ps1 on Windows
• Self-Healing & Adaptive Agents — on failure, the system evolves the graph and redeploys automatically without manual intervention
• Human-in-the-Loop Controls — intervention nodes pause execution for human input with configurable timeouts and escalation policies
• Browser Use & General Compute — native extension support for browser automation alongside general compute tasks
• Model-Agnostic via LiteLLM — supports 100+ LLM providers including OpenAI, Anthropic, Google Gemini, DeepSeek, Mistral, Groq, OpenRouter, and local models via Ollama
• MCP Tool Ecosystem — 102 built-in MCP tools for connecting to CRM, support, messaging, data, file, and internal APIs
• Cost Enforcement & Observability — granular budget controls, real-time cost tracking, spending limits, throttles, and automatic model degradation policies
• Session Isolation & Checkpoint Recovery — checkpoint-based crash recovery and session isolation for production-grade reliability
• Template Agents — click "Try a sample agent" to browse and run pre-built templates or customize them as a starting point