Thoughts
Building a Bot That Can Recover, Not Just Reply
Why XelaBot is shifting from fluent responses toward reliable recovery, approval handling, and operational honesty.
A journal of thoughts
2026
Thoughts
After the Reasoning Trap: Making Agent Claims Unrepresentable
How Render-from-Trace and ActionLedger turn agent honesty from a prompt problem into a system property.
Thoughts
Fixing the Reasoning Trap: Moving XelaBot to Render-from-Trace
How I'm overhauling XelaBot's architecture to eliminate action hallucinations and build trust through deterministic reliability.
Thoughts
The Evolution of My AI Setup: Why I Left OpenClaw for a Custom Vector-Backed Brain
How I evolved my AI setup from OpenClaw to XelaBot, a custom modular agent architecture with semantic memory.
Thoughts
Mein OpenClaw Setup: Ein KI-Assistent auf dem eigenen Server
Wie ich mir mit OpenClaw einen persönlichen KI-Assistenten auf einem VPS eingerichtet habe, erreichbar über Telegram, mit Zugriff auf Kalender, Einkaufsliste, E-Mails und einer geteilten Obsidian Vault als Single Source of Truth.