append-only logging

Traditional agent frameworks treat logs as a post-hoc observability layer, lacking auditability, reproducibility, and efficient branching capabilities. This work proposes ActiveGraph, a runtime system that treats an append-only event log as the sole source of truth, with the working graph serving as its deterministic projection. Agents coordinate by generating new events in response to graph changes, eliminating direct inter-component calls. This architecture enables deterministic replay of arbitrary execution traces, low-cost event-level branching, and end-to-end provenance from high-level objectives down to individual model invocations. By integrating event sourcing, reactive graph computation, typed edge bindings, and LLM-powered behavioral modules, we implement a due diligence case study whose complete causal structure can be reconstructed solely from logs, demonstrating the system’s effectiveness for auditable and forkable agent applications.

Existing autonomous commercial protocols struggle to achieve interoperable, tamper-proof auditing and event temporal verification across heterogeneous domains. This work proposes a verifiable global event timeline architecture that constructs a reproducible, tamper-resistant AI fraud intelligence training pipeline by formalizing event schemas, employing deterministic batching, leveraging Merkle append-only commitments, and anchoring events to blockchain-based timestamps. The approach innovatively integrates cryptographic fraud markers—binding risk labels with anchored evidence—and a data provenance model to establish a verifiable, traceable, AI-ready intelligence layer. Evaluated on a prototype processing 50,000 events, the system constructs Merkle trees in just 47 milliseconds, achieves end-to-end verification in under 0.013 milliseconds, and exhibits logarithmic proof size growth, yielding a 14.4× improvement in verification efficiency over linear scanning.

Databases continuously evolve through operations such as schema changes, version updates, and data transformations; however, existing approaches typically address these functionalities in isolation, lacking a unified abstraction. This work proposes the first integrated model that unifies continuous schema evolution, version management, and data transformation within a single framework. Built upon general-purpose computational primitives, the model supports operation provenance, conditional update propagation, and change alerts, while employing a declarative mechanism to manage the co-evolution of dependent artifacts—including views and machine learning models. A prototype system implements this framework using an enhanced, parameterized Prolly Tree—a Merkle tree–inspired data structure—to construct a relational-like engine. Experimental evaluation demonstrates that the proposed approach is both feasible and offers tunable performance across diverse evolution scenarios.

Current research on agent memory systems is hindered by fragmented architectures, tightly coupled components, evaluation protocols bound to specific datasets, and insufficient support for heterogeneous memory types. This work proposes an interoperable memory research framework that decouples stages of the memory lifecycle through declarative data contracts, separates benchmark datasets from execution protocols, and introduces a unified computational interface to harmonize symbolic, neural, and multimodal memory representations. The framework enables, for the first time, cross-platform plug-and-play memory components, orthogonal separation of evaluation protocols and datasets, and unified runtime coordination of heterogeneous memory types. Experiments demonstrate that the approach facilitates cross-system integration, flexible reconfiguration of evaluation pipelines, and systematic isolation and analysis of memory design variables.

This work addresses the challenges of memory persistence, evolution, and correction faced by lifelong intelligent agents through a novel compute-aware memory plane architecture. The proposed framework models memory as versioned capsules endowed with revision-aware capabilities, enabling unified management of semantic content, provenance metadata, validity status, and reusable runtime states, while supporting replay-aware service integration. A prototype implementation built upon vLLM incorporates persistent capsule states, concurrency-safe invalidation mechanisms, OpenAI-compatible routing bypasses, typed metadata contracts, and a real-time benchmarking framework. Experimental results demonstrate that the approach achieves low latency, high reuse rates, and behavior correctability under prompt-level text retrieval, key-value-aware routing, and cascading failure control scenarios.

Traditional distributed systems struggle to support modern autonomous infrastructures that integrate stochastic models and autonomous agents. This work proposes the Post-Deterministic Distributed System (PDDS) model, introducing for the first time its five architectural pillars. Its core innovation is a "cognitive state replication" mechanism that extends consistency from data visibility to knowledge visibility, alongside a novel fault classification framework. By leveraging protocol-driven development, verifiable agent infrastructure, and semantic quorum guarantees, PDDS enables coordination among semantically equivalent yet executionally divergent agents. This approach achieves verifiable semantic rollback and cross-agent reasoning consistency, establishing a theoretical foundation for trustworthy autonomous systems.

This work addresses the critical challenge that while autonomous agents enhance operational efficiency, their failures can lead to sudden and irreversible consequences. To mitigate this risk, the paper introduces, for the first time, the concept of an “agent data environment,” which reimagines traditional passive data systems by constructing an active execution substrate that integrates files, APIs, applications, and system states. This architecture simultaneously enables capability enhancement and enforces safety constraints through embedded mechanisms for proactive intervention and assurance. By doing so, it not only amplifies agent effectiveness but also effectively bounds the impact of potential failures, thereby establishing a foundational framework for highly reliable autonomous automation.

This work addresses concurrent write conflicts in multi-agent code co-synthesis by proposing the AI-Atomic-Framework (ATM), which integrates task intent, repository scope, write permissions, validation, and evidentiary obligations into an auditable governance chain through a specification-driven mechanism. Key innovations include a content identifier (CID)-mediated coordination protocol for shared writes, adapter-guided semantic atomic decomposition, a virtual atomic mechanism to handle incomplete mappings, and delegation of final writes to a neutral custodian. The approach is empirically validated in a single-domain setting, demonstrating feasibility, auditability, and limited recoverability across twelve scenario configurations, multiple case studies, and an external adoption assessment.

Existing agent frameworks struggle to persist and reuse exploration experience accumulated over long-horizon tasks. This work proposes Trellis, a system that, for the first time, models agent experience as first-class graph database objects, enabling structured, queryable, and governable storage. By integrating vector-guided graph retrieval, materialized views, and time-travel queries, Trellis unifies cross-session experience reuse, crash recovery, horizontal scalability, and closed-loop training within a single architecture. Deployed in Meta’s KernelEvolve production system, Trellis accelerates task completion by approximately 10× while reducing token consumption by 52%.

Current agent memory systems treat writes as immediate truths, rendering them vulnerable to contamination, obsolescence, or incompleteness, which can trigger irreversible errors. This work proposes MemTX—the first transactional belief-commit protocol—that employs snapshot isolation to stage writes and validates them based on evidence, authority, provenance, and validity, committing irreversible actions only when beliefs are deemed trustworthy. MemTX further supports cascading rollback and repair mechanisms. We introduce attribute-based testing and bounded exhaustive verification—covering 5.5 million states—to formally prove the protocol’s correctness. Experimental results demonstrate that MemTX consistently outperforms eight baselines across five backbone models from three model families, achieving significant gains on four backbones, matching performance on one, and crucially inducing zero downstream harm.