Paper: Logos: An Agent Harness on a Cross-Process Bus Authors: Hanzhang Jia, Liheng Zeng, Hao Cheng, Yi Gao, Bo Ma Field: Machine Learning arXiv: 2608.28553
Abstract (translated)
Modern agent systems assemble capabilities at runtime, and this dynamic composition has recently received a complete formal treatment in the spatiotemporal-composability calculus, in which a capability is a component carrying a tracked inverse, and agents are assembled as plugins. This plugin form is carried by a single process sharing one context, a carrier that places all components in one physical failure domain: a fault suspends every component at once, and process death interrupts every session the process hosts.
This paper shows that neither the modeling nor the calculus binds an agent to one process — the statelessness of the language model keeps all cross-step state outside the model, and the soundness invariant is defined on the state space alone. These observations condense into four lemmas whose premises are the calculus's assumptions and the statelessness of language-model inference.
On these lemmas the authors build Logos, a ROS-like cross-process agent framework in which each plugin is a process, and the only shared state is an append-only transcript. With termination injected at four boundaries of the tool-call loop, eighty sessions were recovered without duplicated effects. Compared with a single-process reference configuration under identical faults, one fault interrupts every coexisting session in the monolithic setup, whereas in the peer-process construction a fault terminates at a single node.
Key Takeaways
- Agent capabilities need not be bound to a single process; the underlying calculus permits cross-process decomposition.
- Four lemmas formalize this, relying on the statelessness of LLM inference and state-space-only soundness invariants.
- Logos implements this as a peer-process agent harness with an append-only shared transcript.
- Fault-injection experiments: 80 sessions recovered with no duplicated side effects; faults stay isolated to one node instead of killing all coexisting sessions.