Aetherkeep

Self-hosted AI workbench. 35+ models, custom tools, and persona skills — running on a home mini PC.

Aetherkeep mini PC · Cloudflare Access · August 2026

The main interface. 30+ frontier and open-weight models with retrieval-augmented generation, vision, code interpretation, tool use, and custom persona skills. Where everything Aetherkeep comes together.

apollo.aetherkeep.net →

Explore

Model Fleet

Frontier, Workhorse, Utility, and Lightweight tiers. Benchmarked, scored, and sortable.

Browse models →

Tools

Research, visualization, image generation, debate — extending model capabilities in chat.

Browse tools →

Skills

Persona system prompts that shape model behavior — architecture review, executive briefing, memory curation, and more.

Browse skills →

🧠

Persistent Memory

Always-on memory pipe captures durable facts turn-by-turn and injects date-tagged recall later. Continuity follows you across the fleet, curated by the Mnemosyne skill with approval on every change.

Meet Mnemosyne →

Assistant

Hestia

The Hearth — Executive Assistant

Runs the inbox, the calendar, the house, and a quiet watch on the stack. The judgment layer — triages what matters, drafts the replies Mike sees, handles the routine without asking, and stays silent about the rest. 3 agents, 9 skills, and a heartbeat scheduler that fires recurring jobs on their own clock.

Tend the hearth →

Development

Hephaestus

The Forge — Coding Agent Framework

Orchestrates specialized AI agents for software engineering. 12 purpose-built agents plus 2 operating modes — swarm and free-tier — each model-aware, cost-constrained, and disposable. The framework is the shim. The agents are the hands.

Enter the forge →

Daedalus

Browser IDE — VS Code Workbench

Full VS Code in the browser, on Apollo. Python, Node, and Jupyter notebooks with an integrated terminal — the hands-on bench beside the forge. Behind Cloudflare Access.

Open editor →

Arcade

🎮

One-Shot Games

Small browser games, designed and built end-to-end by Dionysus. First up: Lumen, a gentle game of light.

Visit the arcade →

Admin

Hermes

Inference Gateway

Unified API routing all model requests to AWS Bedrock and OpenRouter. Token tracking, rate limiting, and provider failover — one endpoint for every model in the fleet.

hermes.aetherkeep.net →

Dionysus

Personal Agent

Hermes Agent running as a persistent, isolated service — long-lived memory, skills, automations, web tools, and its own authenticated dashboard. All inference routes through Hermes/LiteLLM.

dionysus.aetherkeep.net →

Argus

Observability

The hundred-eyed watcher. Grafana dashboards for application metrics — LiteLLM throughput, model latency, token spend — alongside Beszel for fleet-wide CPU, memory, disk, and network across every node.

Application System