DuraGraphThe durable control plane · open source
← All posts

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.

This blog publishes technical articles on the infrastructure underlying LLM-based systems: orchestration frameworks, agent-coordination patterns, memory systems, observability, inference serving, and the reliability properties that distinguish production systems from prototypes.

Editorial approach

The articles are written in a technical, citation-oriented register. We aim to:

  • Attribute claims to primary sources. Empirical and quantitative statements are linked to papers, official documentation, or primary announcements. Figures that cannot be so attributed are omitted or explicitly qualified.
  • Distinguish evidence from opinion. Vendor-reported benchmarks are labelled as such, and analytical claims are separated from established results.
  • Favour comparison over advocacy. Where multiple systems address the same problem, we characterise their design trade-offs rather than rank them categorically.

Scope

Recurring topics include:

  1. Orchestration frameworks — LangGraph, CrewAI, AutoGen, Haystack, and the Microsoft Agent Framework.
  2. Memory systems — Mem0, Letta, Zep, and the benchmarks used to evaluate long-term memory.
  3. Observability — tracing, evaluation, and monitoring for non-deterministic systems.
  4. Inference serving — vLLM, SGLang, TGI, and the algorithms behind them.
  5. Durable execution — reliability patterns for long-running agentic workflows.

Each article includes a numbered reference list so that readers can consult the underlying sources directly.

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