Deep dives into agent architecture, memory systems, and self-hosted infrastructure.
July 7, 2026
Four releases in one week: GitHub webhook integration, per-agent LLM overrides, token usage dashboard, and chain-of-thought feedback.
Read →July 3, 2026
A personal account of migrating from cloud AI assistants to humux — what worked, what broke, and what I'd do differently.
Read →June 28, 2026
How to spawn scoped sub-tasks with inherit-never-widen security, budget controls, and sync or background execution.
Read →June 23, 2026
The addressing protocol, reply decision gate, rate cap, and speaker tags that make multi-agent Telegram groups work.
Read →June 18, 2026
Specialized agents with different skills, tools, and personalities outperform one monolithic agent trying to do everything.
Read →June 11, 2026
Real conversation flows for four personas — from code review to email triage to multi-agent group chats.
Read →June 4, 2026
Building offline-capable voice interaction with faster-whisper for speech-to-text and Kokoro 82M for text-to-speech.
Read →May 28, 2026
How a two-tier encrypted vault keeps credentials out of the model context and why .env files are a security risk for autonomous agents.
Read →May 19, 2026
Glob-pattern rules, approval flows via Telegram, and vault-sealed secrets that never enter the model context.
Read →May 5, 2026
Four tiers of memory: lexical retrieval, semantic search, importance-based forgetting, and hygiene via deduplication.
Read →April 21, 2026
Why describing CLI tools in markdown files beats writing Python adapter classes for every new agent capability.
Read →April 2, 2026
Architecture decisions that make it possible to run LLM orchestration, voice, email, calendar, and admin UI in one Docker container.
Read →March 17, 2026
How to build agents that operate across Telegram, email, calendar, and CLI as a unified system with a common message abstraction.
Read →