archive
Every paper Pith has read. Search by title, abstract, or pith.
128 papers in cs.OS · page 1
-
C++ runtime runs embodied models on any robot hardware
Embodied.cpp: A Portable Inference Runtime of Embodied AI Models on Heterogeneous Robots
-
GNSS pipeline timing becomes deterministic with logical time model
Ensuring Deterministic Timing in a Federated GNSS Correction Pipeline with Lingua Franca
-
Workflow combines CPU and GPU TEEs for private AI
EnclaveX: End-to-End Confidential AI with CPU/GPU TEEs
-
Accessibility metadata powers AI agents without screenshot reliance
LUMOS: A Semantic Operating-System Layer for Accessibility-Grounded AI Agents
-
Kernel engine enforces agent policies on paths tool calls miss
ActPlane: Programmable OS-Level Policy Enforcement for Agent Harnesses
-
OS kernel engine enforces agent policies beyond tool calls
ActPlane: Programmable OS-Level Policy Enforcement for Agent Harnesses
-
Kops adds native ops to eBPF JIT via verified proof sequences
Kops: Safely Extending the eBPF Compilation Pipeline with Native Operations
-
Hierarchical CXL+RDMA pool speeds MicroVM restores 2.2x
Aquifer: Hierarchical Memory Pooling with CXL and RDMA for MicroVM Snapshots
-
Master core LMS regulator cuts memory slowdowns over Memguard
LMS-AR: LMS Prediction-based Adaptive Regulator for Memory Bandwidth in Multicore Systems
-
Android harness lifts AI agent success rate by 21 percent
AOHP: An Open-Source OS-Level Agent Harness for Personalized, Efficient and Secure Interaction
-
Decoupled permissions speed secure mobile LLM inference 10x
FlexServe: A Fast and Secure LLM Serving System for Mobile Devices with Flexible Resource Isolation
-
TrustZone secures mobile LLMs with 10x faster first-token time
FlexServe: A Fast and Secure LLM Serving System for Mobile Devices with Flexible Resource Isolation
-
On-device LLMs cut energy use up to 65% by lowering NPU speed
EnerInfer: Energy-Aware On-Device LLM Inference
-
Reverse engineering maps Apple Neural Engine internals
Apple Neural Engine: Architecture, Programming, and Performance
-
OS reframes as intent filter for AI agents
AgenticOS: An Intent-Oriented Secure Operating System Architecture for Autonomous AI Agents
-
OS takes over LtRAM management with read-only app pages
The Kernel's Write: Application Read-Only Memory
-
Four OS subsystems enable full-fidelity cluster simulation
LiveStack: OS Support for Cluster-Scale Full-Stack Live Simulation
-
CloakLM fragments GPU weight memory to block exfiltration
CloakLM: Obfuscating GPU Memory Layout to Mitigate Model Ex-filtration for Serving
-
Cordon wraps LLM tool calls in semantic transactions for pre-commit validation
Cordon: Semantic Transactions for Tool-Using LLM Agents
-
Client-server system keeps language model music synced to live signals
Real-Time Language Model Jamming: A Case Study for Live Music Accompaniment Generation
-
Model virtualization shares one backbone to host 6x more tasks
FMplex: Model Virtualization for Serving Extensible Foundation Models
-
Middleware gives microcontrollers per-container scheduling and access control
TinyContainer: Container Runtime Middleware Enabling Multi-tenant Microcontrollers with Built-in Security
-
Datalog language adds dynamic states to access policies
Policy Description Language for Authorization using Logic-Based Programming
-
TOMOYO Linux bases access control on app execution history
TOMOYO Linux: A Mandatory Access Control Method Based on Application Execution State
-
AgileOS virtualizes CUDA to protect GPU services
AgileOS: A GPU Operating System Layer for Protected CUDA Services
-
Keeping old servers can cut total carbon versus upgrades
CarbonSim: A Lifecycle-Aware Framework for Evaluating Carbon Tradeoffs in Hardware Upgrade Decisions
-
GPUs bypass CPU for 3.2x faster remote flash access
GNStor: Design of GPU-Native High-Performance Remote All-Flash Array
-
Runtime primitives replace tool calls as trust boundary for LLM agents
Agent libOS: A Runtime Substrate for Capability-Controlled Self-Evolving LLM Agents
-
LibOS lets LLM agents evolve while blocking unauthorized side effects
Agent libOS: A Runtime Substrate for Capability-Controlled Self-Evolving LLM Agents
-
Metastable failures trace to destabilizing component cycles
Characterizing Metastable Faults and Failures
-
Relative idleness ranking lifts agent throughput 20-71%
Idleness is Relative: Exploiting Tool-Call Idle Windows for Offloading in Agentic Systems with MORI
-
Adaptive controller cuts deadline misses in robot edge offloading
Edge-Based QoS-Aware Adaptive Task Placement: A Closed-Loop Control in Multi-Robot Systems
-
Kernel fuzzer now sees argument values at function calls
Beyond Edge Coverage: Per-Task Data-Flow Extraction at Kernel Function Boundaries via LLVM
-
RTP-LLM speeds LLM loading by 4.7x-6.3x and cuts latency 35%
RTP-LLM: High-Performance Alibaba LLM Inference Engine
-
IORM manages I/O for thousands of consolidated databases without interference
IORM: Hierarchical I/O Governance for Thousands of Consolidated Databases on Oracle Exadata
-
Mediator validates signed manifests for safe unprivileged promotion
A Secure, Manifest-Based Framework for Delegated Privilege Promotion
-
Hotpatching cuts automotive update time to microseconds
Patchlings: Safety-Preserving Flash-Based Hotpatching for Automotive Microcontrollers
-
Batched lock matches FIFO wait bounds while favoring high-priority tasks
Bounded Priority-Aware Locking for Real-Time Kernels
-
Sandlock confines AI code with unprivileged Linux primitives
Sandlock: Confining AI Agent Code with Unprivileged Linux Primitives
-
Perceptron eviction beats FIFO by up to 10% in Linux cache
LearnedCache: An eBPF-Integrated Perceptron-Based Eviction Policy for the Linux Page Cache
-
DeltaBox cuts sandbox checkpoint to 14 ms for AI agents
DeltaBox: Scaling Stateful AI Agents with Millisecond-Level Sandbox Checkpoint/Rollback
-
DeltaBox cuts AI agent checkpoint and rollback to 14 ms and 5 ms
DeltaBox: Scaling Stateful AI Agents with Millisecond-Level Sandbox Checkpoint/Rollback
-
Multicore interference stall bounded by (N-1) memory latencies
A Per-Access Upper Bound for Shared-Resource Interference in Direct-Mapped Multicore Architectures
-
ParaCell cuts secure container latency up to 88%
ParaCell: Paravirtualized Secure Containers with Lightweight Intra-Container Isolation and Intent-Driven Memory Management
-
EFQ gives mobile tasks proportional power shares
Energy-Based Fair Queuing Scheduler for Mobile Systems
-
Managed runtime extension cuts CXL slowdown by 22-84%
Clove: Object-Level CXL Memory Management in Managed Runtimes
-
SSV lifts LLM inference speed up to 3.49x
SSV: Sparse Speculative Verification for Efficient LLM Inference
-
SpecSA resolves mismatch to speed long-context LLM inference
SSV: Sparse Speculative Verification for Efficient LLM Inference
-
C2CServe cuts LLM cold-start latency up to 7.1x on GH200
C2CServe: Leveraging NVLink-C2C for Elastic Serverless LLM Serving on MIG
-
LLM semantics from names predict load phases and cut storage overloads 79%
TIDAL: Recovering Temporal Phase for Cloud Block Storage Placement from LLM-Derived Semantics