Automation2026
Oreo & Hermes Agent Runtime
The operations agent that runs my business loops on a schedule
Results
- ✓One reasoning path instead of a different one per script
- ✓Every action it takes is written down and can be read back
- ✓The gated actions have never fired without a human saying yes
The Situation
- •Loops for outreach, reporting, and alerting were separate scripts with separate habits
- •Each one asked for a decision in its own way, so behaviour drifted between them
- •An agent that can act needs a record of what it did and a hard stop where humans belong
The System
- •Oreo runs the operations loops on a schedule
- •Hermes is the reasoning layer, reached through one adapter so every system asks the same way
- •Human gates: sending, spending, and publishing stop and wait for approval
- •Everything runs locally and writes an audit log
How it runs
TriggerSchedule firescron
ActionOreo runs the loopOutreach, reporting, alerts
ActionHermes decidesOne reasoning adapter
DecisionHuman gate?Send, spend, publish
ActionTakes the actionOnly when allowed
OutputAsks me firstApproval lane
OutputAudit logEvery action, always
Stack
Pythonlocal modelscronaudit log
What's Next
- Keep the gate list short and explicit
- Retire any loop that stops earning its schedule
Want similar systems?
I build AI systems for private practices and local service businesses in the DMV, and cinematic websites that convert. Start with a free AI business audit.
Next projectBlog Publishing Automation