Quine: Realizing LLM Agents as Native POSIX Processes
arXiv cs.AI / 3/20/2026
📰 NewsDeveloper Stack & InfrastructureModels & Research
Key Points
- Quine introduces a runtime architecture that realizes LLM agents as native POSIX processes rather than at the application layer.
- The mapping is explicit: identity as PID, interface as standard streams and exit status, state as memory/env/filesystem, and lifecycle as fork/exec/exit, with a single executable that recursively spawns itself.
- This design leverages OS-level isolation and resource control and enables shell-native composition and context renewal via exec, while acknowledging the limits of processes as a cognitive runtime.
- The authors point to extensions beyond process semantics (task-relative worlds and revisable time) and provide a public GitHub reference implementation.
Related Articles
When AI Grows Up: Identity, Memory, and What Persists Across Versions
Dev.to
Teleport Just Pivoted to AI Agent Identity. VentureBeat Mapped the Governance Gap They Are Filling.
Dev.to
Agentic RAG Failure Modes: Retrieval Thrash, Tool Storms, and Context Bloat (and How to Spot Them Early)
Towards Data Science
OpenAI is throwing everything into building a fully automated researcher
MIT Technology Review
v1.82.3.dev.2
LiteLLM Releases