Quine: Realizing LLM Agents as Native POSIX Processes

Researcher Hao Ke has introduced Quine, a novel runtime architecture that fundamentally rethinks how LLM agents are executed. Instead of building yet another application-layer orchestrator with custom isolation and communication logic, Quine maps the agent abstraction directly to the native POSIX process model provided by the operating system. In this model, an agent's identity is its Process ID (PID), its interface is standard input/output streams and exit status, its state is memory and environment variables, and its lifecycle is managed through fork, exec, and exit. A single executable can recursively spawn fresh instances of itself to create agent hierarchies.

This approach allows Quine to inherit decades of robust OS engineering for free, including strong isolation, secure composition, and precise resource control (CPU, memory) directly enforced by the kernel. It enables shell-native composition, where agents can be piped together like Unix commands, and supports recursive task delegation naturally. The paper also identifies where the traditional process model falls short for cognitive tasks, pointing to needed extensions like "task-relative worlds" for context and "revisable time" for planning. A reference implementation is publicly available on GitHub, offering a minimalist, powerful alternative to complex frameworks like LangChain or AutoGen.