All projects
In DevelopmentResearch / Infrastructure
AI Homelab & Agent Experimentation
A personal research environment for self-hosted AI tools, local development, agent frameworks, infrastructure, and practical automation experiments, with a running lab journal.
Last updated Oct 17, 2026
Problem
Problem being solved
Reading about agent frameworks and self-hosted AI is not the same as running them. A dedicated lab makes it possible to test ideas cheaply, measure real trade-offs, and write down what actually works.
Architecture
- 01Agent frameworksorchestration experiments
- 02Local + hosted modelscost · latency · capability trade-offs
- 03Tools & integrationsMCP servers · APIs · automations
- 04Lab journalfindings published as notes
Working experiments vs proposed ideas
| Item | Type |
|---|---|
| Agent assistant with tools, scheduled jobs, and chat integrations | Working experiment |
| Content and media automation pipelines | Working experiment |
| MCP server connections to everyday tools | Working experiment |
| Local model evaluation on Apple silicon | Proposed |
| Multi-agent orchestration patterns | Proposed |
| Dedicated always-on lab server | Proposed |
Lab journal
- 2026-10: Connected an agent to messaging, scheduled jobs, and platform APIs through official developer programs. Learned that most of the work is in OAuth setup, token refresh, and safe approval steps before anything is published.
Future ideas
- A small evaluation harness: same task, several models, measured cost and quality
- Shared tool registry between experiments (see the Modular MCP Tool Platform project)