References

External sources the graph draws on, with publication metadata.

• [[A-MEM Agentic Memory (Xu et al., 2025)]] -- A-MEM proposes agentic memory for LLM agents organized as a Zettelkasten-inspired interconnected knowledge network with dynamic indexing and linking, achieving 85-93 percent tok...
• [[ClawVM Harness-Managed Virtual Memory (Rafique and Bindschaedler, 2026)]] -- ClawVM proposes harness-layer enforcement of atomicity, persistence, and capability boundaries for stateful tool-using LLM agents, arriving independently at the substrate-proper...
• [[Compound AI Systems Optimization Survey (Lee et al., 2025)]] -- A survey of optimization methods, challenges, and future directions for compound AI systems
• [[DeepContext.com Graph (Allen, 2026)]] -- The seed Deep Context graph; donor of the meta-layer (Form Contracts, Predicates, generic graph-operation Skills) grafted into eOS-DeepContext at instantiation
• [[DSPy Compiling Declarative Language Model Calls (Khattab et al., 2023)]] -- DSPy is the foundational paper for treating LM pipelines as text-transformation graphs
• [[Mem0 Production Memory (Chhikara et al., 2025)]] -- Mem0 is a production-oriented memory layer for AI agents using LLM calls to extract facts from conversations, storing them in hybrid vector and graph databases
• [[MemGPT Towards LLMs as Operating Systems (Packer et al., 2023)]] -- MemGPT is the foundational paper framing agent memory as an OS-shaped concern, treating the LLM context window as physical memory and external storage as disk with virtual-memor...
• [[MemMachine Ground-Truth-Preserving Memory (Wang et al., 2026)]] -- MemMachine stores raw conversational episodes and indexes them at sentence level, minimizing LLM dependence for routine memory operations and preserving factual integrity using...
• [[Memoria Modular Memory Framework (Sarin et al., 2025)]] -- Memoria is a modular memory framework that augments LLM-based conversational systems with persistent, interpretable, and context-rich memory, integrating dynamic session-level s...
• [[Meriadoc (ChatTheatre, 2026)]] -- The refererence implementation's sandboxed code-load DSL: source-as-property in the persistent image, multi-phase signal-triggered execution, decoration-and-compile over a denyl...
• [[Model-Harness-Fit (Bustamante, 2026)]] -- Argues the matched pair of model and harness is the proper unit of analysis
• [[Recursive Language Models (Zhang et al., 2025)]] -- Recursive Language Models treat the prompt as an external Python REPL environment the LM examines, decomposes, and recursively calls itself over
• [[Recursive Language Models Paradigm of 2026 (Prime Intellect, 2026)]] -- A Prime Intellect blog post arguing RLMs are the paradigm of 2026
• [[Self-Reflective Language Models (Alizadeh et al., 2026)]] -- Self-Reflective Language Models extend RLM's recursive scheme with uncertainty-aware program-trajectory selection in the same Python REPL environment, achieving 22 percent accur...
• [[Towards Resource-Efficient Compound AI Systems (Chaudhry et al., 2025)]] -- Proposes a declarative workflow programming model and an adaptive runtime system for dynamic scheduling and resource-aware decision-making in compound AI systems
• [[Why Atomicity Matters to AI ML Infrastructure (Borrill, 2026)]] -- A paper arguing atomicity must be explicitly constructed as protocol convergence in AI/ML training infrastructure rather than assumed as a temporal property
• [[Why I Built DSPy Agent Skills (codeandcontext.ai, 2025)]] -- A blog post arguing that procedural knowledge differs fundamentally from raw information availability for agents, framing GEPA optimization plus RLM long context plus procedural...

17 nodes in this section.