MANDAG
2026-10-26

Too many projects, too many ideas, too few hours — one learning a day anyway

Hermes setup

An always-on local AI orchestrator with long-term memory.

The operating environment behind everything else: a resident orchestrator that routes tasks to LLM workers, runs scheduled agents (like the weekly Field Monitor that feeds this site’s news), and maintains long-term memory — nightly ingestion of work sessions and notes into a searchable store. The design principle: the orchestrator plans and verifies; disposable workers execute.

Its panel is where the small watches surface. One of them reads my public repos once a day and says what is new since yesterday — a star, an issue, a question somebody asked and is waiting on. Publishing something is easy to forget afterwards.

HERMES — THE HOUSEHOLD ORCHESTRATOR CHAT — me, from anywhere messages become tasks; answers come back to the same conversation SCHEDULED JOBS watchdogs, monitors, nightly ingest — scripts first, the model only when drift is found THE ORCHESTRATOR — always on routes each task to the cheapest model that can carry it, falls over to the next provider when one runs dry — free first, frontier last every session, every cost everything worth keeping STATE LEDGER who ran, what it cost, which provider — per model, per day; a dashboard reads it so the bill is never a surprise MEMORY BANK nightly ingest: coding sessions and notes become recallable memories — local, embedded, no cloud recall, every task THE POINT OF THE LOOP tomorrow's task starts with yesterday's context — the house assistant that remembers THE CLOUDY BRAIN DWG NO. 004 HERMES SETUP REV. A drawn by: human + machine SHEET 1 OF 1
CODE
  • AI lane limits and live rows read from the registry
  • Portal agents card shows every Foreman lane
  • One shared lockd gate for every Hermes job
  • Hindsight memory moved to plexmaster
  • Foreman becomes the primary work status in portal
  • Hermes portal scaffolded with Hindsight and Obsidian
POSTS