FIG 18.0 · Blog

Notizen vom Baueines Betriebssystems.

Technische Entscheidungen, Architekturdebatten und gelegentlich etwas, das wir falsch gemacht haben.

MCP7 August 2026· Auf Englisch

A complete guide to the Model Context Protocol

What MCP is, how the transport and primitives actually work, how servers are built and connected, and what you are granting when you connect one.

Product7 August 2026· Auf Englisch

On-device speech recognition, and why voice assistants stayed useless

Voice is the fastest input humans have, and we use it for timers. The reason is not accuracy. It is that every mainstream assistant points a microphone at somebody else's data centre.

Security6 August 2026· Auf Englisch

The agentic AI threat model

Why an agent's capabilities are its attack surface, how prompt injection turns tool access into a real exploit, and which mitigations actually hold under pressure.

Architecture5 August 2026· Auf Englisch

Immutable Linux, explained

What a read-only root filesystem actually changes, how A/B atomic updates and BTRFS snapshots work, and why the model suits agentic workloads in particular.

Guide4 August 2026· Auf Englisch

Running AI agents locally: a practical guide

What hardware you need, how quantisation and context length drive memory, which model sizes suit which work, and where local inference stops being the right answer.

Architecture14 July 2026· Auf Englisch

Why an agentic AI needs to be an operating system

Every guarantee that matters for agentic AI (isolation, egress control, rollback, attribution) is enforceable at the OS layer and merely promised anywhere above it.

Product2 June 2026· Auf Englisch

Local-first is a capability argument, not a privacy one

People withhold context from cloud assistants, correctly. That withholding is what makes those assistants mediocre.

Verwandt

Produkt

Editionen

Lösungen

Entwickler

Unternehmen

Sprache auswählen

Amerika

Europa

Naher Osten

Asien-Pazifik