AI Agent
Architecture, tool use, memory, RAG, evaluation, permissions and runtime governance.
Look High. Build Low.
XBSTACK continuously documents practical AI Agent, MCP, LangGraph and Workflow engineering, alongside useful developer tools, long-term research and outdoor field notes.
⌕ Search errors, engineering problems, tools, topics or articles…Architecture, tool use, memory, RAG, evaluation, permissions and runtime governance.
Servers, transports, OAuth, permissions, security and protocol troubleshooting.
State, checkpoints, HITL, failure recovery, concurrency and observability.
n8n, self-hosting, webhooks, queues, retries, idempotency and cost governance.

OpenAI Zero Data Retention does not make an AI agent stack stateless. This guide maps ZDR across Responses API, remote MCP, prompt caching, background mode and Data Residency.

In n8n 2.35.3 Edit Fields, $json.data.sort(), splice(), fill() and copyWithin() return null. Compare official 2.34.5 vs 2.35.3 Docker runs and tested copy-first workarounds.

n8n HTTP Request can return _readableState instead of JSON with Raw Body + explicit JSON Response. See the four-case repro, source path and tested workarounds.

A personal view of why Guizhou attracts travelers, and why first-time visitors may enjoy it more by choosing fewer stops and leaving room for weather, local life and the mountains.

DeepSeek V4-Pro-0813 explained: API pricing, 1M context, 384K output, concurrency, OpenAI/Anthropic setup, and DeepSeek Harness installation.

AI SDK 7.0.66 repro: streamObject can leave result.object and related promises pending after provider errors even when fullStream is consumed. Includes a tested failure fence.