XBSTACK is my long-term personal home, not just an AI troubleshooting site. I build independent products, document engineering decisions, read and rethink ideas about investing, and make room for travel and outdoor life. Search helps new readers find the work; the person and the projects are why it belongs together.
Local-first personal AI memory: product direction, document retrieval, cited answers and current launch status.
Sleep and recovery app in preparation; no public release claim.
See the applications, not just technical demos.
Reproducible Agent, MCP, memory and security engineering.
Long-form reading notes, original judgments and their limitations.
Outdoor experience, photographs and personal observations.
Project cooperation, technical discussion and corrections.
The founding purpose is to build a durable personal body of work. Traffic matters because it connects readers to useful work, but search rankings alone cannot express a developer’s decisions, long-running projects, reading, investments and life.
I prefer to distinguish what I have built and measured from what I have only researched or plan to test. I do not turn every new model, bug or tool into a separate product or SEO page.
Recalia is an independent, local-first personal AI memory product and Lunest is a sleep and recovery product. Both are in pre-release preparation as described on their own pages; neither should be presented as an already available store download.
Individual utilities, such as the Local AI Model Fit tool, live under Tools. Engineering experiments belong under AI; long-running reading, investing, personal observations and outdoor accounts retain their own editorial purpose.
Technical articles are expected to distinguish official documentation and primary evidence from XBSTACK engineering interpretation. When a problem can be reproduced locally, the preferred format includes environment details, minimal reproduction, observed behavior, workaround boundaries and links back to upstream evidence.
Existing URLs are updated when the search intent is the same instead of creating multiple pages that compete for the same query. Related guides are organized into hubs such as AI Agent Security, LangGraph and MCP so readers can move from a broad problem to a specific implementation detail.
Chinese remains the default language. Reviewed English pages use separate /en/ URLs with their own canonical tags and reciprocal hreflang links.
English content is reviewed for structure, terminology, links and factual meaning before it is treated as publicly publishable. Corrections, reproducibility reports and technical feedback can be submitted through the contact page.
