Blog
Essays on
what we build.
Cryptography, AI agents, systems design, and the future of personal assistants. No marketing. Just engineering and opinions.
Why MONO is built in Go instead of Python
95% of open-source AI agents are written in Python. MONO is in Go. The reasons: latency, concurrency, static binaries, and a memory model that doesn't collapse under real load.
Skill vs Tool: the distinction that makes MONO different
In most agents, 'skill' and 'tool' are synonyms. In MONO they're different concepts. That distinction is why a MONO can do 83 things without getting confused.
How MONO learns from you without sending your data anywhere
AI assistants need memory to be useful, but storing your info on a central server is a security bet. Per-user local memory, knowledge graph on your VPS.
How MONO decides what to do: anatomy of the router
When you write 'book me with the dentist Friday 4pm', your MONO picks which of its 47 skills to use in 200ms. How the router is designed and why speed matters.
Your own server: why MONO isn't SaaS
Most AI assistants run on a shared server where your data lives next to thousands of others. MONO doesn't. Every account gets its own isolated server. Why it matters.
ML-KEM-768 explained: why post-quantum encryption is no longer optional
By 2030, a quantum computer will break RSA-2048 in hours. Every encrypted message today can be decrypted tomorrow. Why MONO ships ML-KEM-768 (NIST FIPS 203).
How I built an AI agent with 83 skills in Go — solo
8 months, 47 skill packages, 33 Dynamic UIs, SurrealDB 3.0, a Haiku router that picks which skill to run, and an Intelligence Engine v4 that talks on WhatsApp. Lessons from the trenches.
WhatsApp as OS: why the future isn't apps
Installing apps is friction. Logging in is friction. Learning new UIs is friction. The future converges on a single text thread where you talk and things happen.