Agentic NixOS: Building a Safe Control Layer

A six-part series on building Agentix, a cautious agent-control layer for NixOS. From philosophy to MVP to roadmap.

What interface forms are appropriate when agency is distributed across humans, models, and protocols?

• 1. I. Why Agentic NixOS Needs a Safety Layer
  The motivation: NixOS is declarative and reproducible, but that does not make it safe for LLM operation without a control layer.
  Apr 19, 2026 · Systems
• 2. II. Building Agentix: A Proposal-First Control Layer
  The architecture: proposal-first workflows, goal runner, and verification without switching.
  Apr 21, 2026 · Systems
• 3. III. Dirty Trees, Stale Patches, and Git as a Safety Boundary
  Guardrails around boring failure modes: dirty trees, stale patches, and proposal lifecycle.
  Apr 23, 2026 · Systems
• 4. IV. JSON Outputs, Audit Logs, and Preparing for LLM Operators
  Structured output and audit trails: the foundation for safe automation.
  Apr 25, 2026 · Systems
• 5. V. Agentix v0.1 MVP: What Works Now
  The full MVP feature list: what is included and what is intentionally left out.
  Apr 27, 2026 · Systems
• 6. VI. What Comes Next: From Human-in-the-Loop to Safe Autonomy
  The roadmap: sandbox-first autonomy, worktrees, VMs, and what "agentic" should actually mean.
  Apr 29, 2026 · Systems