About Agentos
AgentOS is an open source TypeScript runtime for autonomous AI agents, distributed as an npm package from Framers. It combines graph based orchestration, a cognitive memory system, streaming guardrails and a voice pipeline in one library, with support for 11 LLM providers.
It is aimed at TypeScript developers building agents that need to remember, collaborate and extend themselves. The memory system is modeled on cognitive science: memories are encoded with a strength that decays over time along a forgetting curve, grows with successful retrieval and is rewritten through consolidation, and sessions keep conversation history separate for each user. An optional HEXACO personality profile tunes an agent's tone and memory behavior. Runtime tool forging lets an agent write a new TypeScript function when no existing tool fits; a separate LLM judge approves it and it runs in a sandbox before joining the catalog. Other features include six multi agent orchestration strategies, multimodal RAG, media generation, deep research, voice IVR and workspace files that define an agent's identity and limits. The project reports 85.6 percent on the LongMemEval S memory benchmark and 70.2 percent on LongMemEval M.
AgentOS is free under the Apache 2.0 license. No hosted service or paid plan is listed; costs come from the model providers you connect.
Documentation, recipes, an extension catalog and benchmark write ups are on the project site, with community support through a Discord server.