Local AI, Hermes and GPT-6 Privacy Guide

What Local AI, Hermes and GPT-6 Privacy Guide does

This repository is a guide, written mainly in Thai with an English summary, to designing an AI workflow where data stays on your machine by default. Hermes Agent orchestrates, a local privacy service handles document access, PII detection, masking and retrieval, and normal work goes to a local OpenAI-compatible model endpoint. Only a hard task may cross a gate, and then only as minimal masked context shown to the user for approval, sent to the OpenAI Responses API and the model the README calls GPT-6 Astra.

The README lists five principles: local-first storage, local data masking, minimal context, explicit approval before every egress, and no silent cloud fallback. It stresses that masking reduces risk without guaranteeing anonymization and that scoped tools, allowlists and approval are defense in depth, not containment. It points to the Hermes SECURITY.md, which says OS-level isolation is the real boundary. The repo has architecture, privacy and Hermes docs plus an offline dry-run payload example, and no installed local model.

Key features

• Architecture write-up for a local-first hybrid AI workflow with Hermes as orchestrator
• Document on data masking with stable placeholder tokens kept in a local token map
• Doc on why Hermes suits agent workflows, with a reminder that it does not make data safe by itself
• Offline example that builds a masked Responses API payload without calling the API
• Five design principles including explicit approval and no silent fallback
• English summary alongside the Thai text

When to use it

• Designing a privacy boundary around Hermes Agent for sensitive Thai-language documents
• Reviewing what stays local and what may reach a cloud model
• Checking a masked payload before any API call

Who it is for: Thai-speaking developers and teams planning a local-first Hermes setup that handles private documents.

How it fits with Hermes Agent

A written guide that places Hermes Agent in the orchestration layer of a local AI design, with security caveats drawn from the Hermes security policy.

Note: The README says it is a principles guide, not a production security control, and the repository contains documentation and an offline example rather than a running local model.

FAQ