Hermes Brain

Hermes Brain is a public, zero-knowledge template for a modular agent skeleton. The goal is to share the architecture and guardrails without publishing private history, memory, or tuned behavior.

What this template contains

• A minimal core loop in brain.py
• Neutral runtime and privacy defaults in policies/
• Example capability scaffolding in capabilities/
• Sample inbox input and an empty outbox
• A placeholder memory directory for private deployments, including an initial outcomes.jsonl file for future append-only memory

Public vs. private

This repository intentionally separates two layers:

• Public template: clean structure, guardrails, and reusable capability hooks
• Private runtime: local memory, environment-specific tuning, and learned behavior

The public version should be useful on its own, but it should not carry the accumulated knowledge of a private deployment.

Modes of use

1. Template mode (default)

• memory/ is treated as private state
• each agent learns independently
• the repository stays clean and portable

2. Sync mode (advanced)

• memory/ is committed and shared intentionally
• agents can inherit collective experience
• this is useful when you want multiple instances to stay aligned

Quick start

python3 brain.py

This processes any JSON files in inbox/ and writes a decision report to outbox/decisions.json.

Repository layout

brain.py
capabilities/
policies/
inbox/
outbox/
memory/
README.md
VERSION

Suggested next steps

• Adjust policies/runtime_rules.json for your environment.
• Add new capabilities under capabilities/.
• Keep memory/ empty in the public repo or use it as a deliberate sync surface in advanced deployments.
• If you adopt append-only memory, write one JSON object per line to memory/outcomes.jsonl.