Blog
Thoughts on technology, enterprise architecture, and lessons from two decades of building solutions.
Showing 1–5 of 16 posts
The story of the wiki both humans and agents can use We still pick Obsidian for quick notes — but in the agent era docs are created at 100x speed by TUI agents on a remote box. A story of the wiki we borrowed, the failover we tuned, and the dynamic spac Read article Why I Built a Local AI Agent Hub (And Why You Might Want One) Cloud agent APIs are expensive, rate-limited, and leak your data. Here's the local-first alternative I run on a homelab: one Docker MCP gateway, a dozen tool servers, any local LLM — and what breaks along the way. Read article Use Case Deep Dives: Four Walk-Throughs You Can Replicate Four end-to-end local AI agent pipelines on the MCP gateway: academic research downloads-to-knowledge-base, multi-source web research with screenshots, document conversion, and multi-day orchestrated research. Read article Connecting Your Local LLM to MCP Your LLM speaks OpenAI function calling, not MCP. Bridging Ollama and LM Studio to a Docker MCP gateway — the client code, the context-window math for tiers of tools, and the connection issues you'll hit. Read article Lessons Learned, Gotchas, and What's Next Six weeks of running local AI agents on a Docker MCP gateway, distilled: context-window math, volume-path perspectives, the colon separator, longLived discipline, and the roadmap from Neo4j to multi-agent swarms. Read article