DuraGraphThe durable control plane · open source
The Logbook

Notes on durable AI infrastructure.

RSS feed ↗
  • December 26, 2025

    A Survey of AI Agent Orchestration Frameworks

    A comparative survey of LangGraph, CrewAI, AutoGen, and Haystack, examining their orchestration models and appropriate use cases.

    agent-frameworkscomparisonorchestration
  • December 26, 2025

    About This Blog

    An introduction to this blog's scope and editorial approach—technical, citation-oriented writing on AI orchestration, agent frameworks, and LLM infrastructure.

    announcementduragraph
  • December 12, 2025

    AI Infrastructure in 2025: A Retrospective

    A retrospective on developments in AI infrastructure over 2025—durable execution, production memory systems, framework consolidation, observability, and LLM gateways—with references to primary sources.

    year-in-reviewtrends2025predictions
  • November 22, 2025

    LLM Serving at Scale: PagedAttention, RadixAttention, and Production Inference

    Inference throughput and memory efficiency increasingly determine the cost of production LLM systems. This article surveys vLLM, SGLang, and TGI and the algorithms behind them.

    inferencevllmsglangperformance
  • November 8, 2025

    Coordination Patterns for Multi-Agent Systems

    A structured account of four coordination patterns for multi-agent LLM systems—hierarchical, sequential, debate, and reactive—together with their failure modes and durability requirements.

    multi-agentorchestrationpatternsproduction
  • October 25, 2025

    A Comparative Survey of LLM Gateways: LiteLLM, Portkey, Kong AI, and Bifrost

    LLM gateways provide failover, caching, rate limiting, and cost control between applications and model providers. This article compares four gateways on architecture and reported performance.

    gatewaylitellmportkeyinfrastructure
  • October 11, 2025

    Observability for LLM Agent Systems: Tracing, Evaluation, and Debugging

    Effective operation of LLM agents depends on observability: the ability to trace, evaluate, and debug non-deterministic systems. This article surveys the tooling landscape and the metrics it exposes.

    observabilitylangfusemonitoringproduction
  • September 27, 2025

    Microsoft Agent Framework: Consolidating AutoGen and Semantic Kernel

    Microsoft has consolidated AutoGen and Semantic Kernel into a single Agent Framework. This article examines the merger, its stated design goals, and the trade-offs it presents for production teams.

    microsoftautogenenterpriseagents
  • September 12, 2025

    Memory-Augmented Agents: A Survey of Production Memory Systems

    Persistent memory extends LLM agents from stateless responders to context-aware systems. This article surveys three production memory architectures and the benchmarks used to evaluate them.

    memoryagentsmem0lettaproduction
  • August 28, 2025

    Durable Execution for AI Workflows: Applying Temporal Patterns to Agentic Systems

    As LLM agents move from prototypes to production, reliability requires more than retry logic. This article surveys durable-execution patterns and their application to long-running agentic workflows.

    durabilitytemporalarchitectureproduction
DuraGraph is the open-source AI workflow control plane, built for production. Self-hosted and event-sourced.
Apache 2.0 · © 2026 DuraGraph · Privacy · Cookies