AgentOS (agentos.sh)

AgentOS is an open-source framework written in TypeScript designed for building production-ready autonomous AI agents. Developed by Frame.dev, it targets developers seeking to create adaptive agents capable of remembering, adapting, and generating new tools during runtime. The framework emphasizes emergent intelligence and supports multi-agent orchestration, enabling multiple agents to collaborate and dynamically spawn specialists when required by the task.

A distinguishing feature of AgentOS is its cognitive memory system, which is grounded in neuroscience concepts such as Ebbinghaus decay and reconsolidation. This allows agents to recall and utilize information over time, contributing to their adaptability. The framework also includes multimodal retrieval-augmented generation (RAG), AI guardrails that are GDPR-ready, and voice pipeline capabilities. AgentOS supports integration with 11 large language model (LLM) providers, offering flexibility in building agents that can leverage different AI backends. Agents built with AgentOS can write their own TypeScript functions mid-task, which are then reviewed and approved by a judge agent, and new specialist agents can be spawned at runtime to expand the system’s capabilities.

For development and management, AgentOS offers a desktop application called AgentOS Workbench. This tool provides a visual agent builder with drag-and-drop workflows, real-time testing and debugging tools, and options for one-click deployment to either cloud or local environments. The framework is self-hosted, allowing deployment on the user’s own infrastructure, and is distributed under the Apache 2.0 license for the core, with MIT licensing for agents, extensions, and guardrails.

AgentOS has demonstrated its effectiveness with benchmark results such as 85.6% on LongMemEval-S, and it is used as the foundation for Paracosm, an AI agent swarm simulation engine that enables scenario-based simulations with AI commanders exhibiting distinct personalities. Extensive documentation, code examples, and community support are available through official channels, including Discord and GitHub.