Hermes Agent — Multi-Service AI Integration

Multi-service AI agent setup: Hermes Agent + Polza AI + Telegram bot + voice commands + Holographic memory system.

What's Inside

.hermes.md

Holographic memory rules for the agent — fact storage structure, trust scoring, search strategies, and maintenance schedules. Defines how the agent remembers user preferences, project context, and tool configurations across sessions.

RULES.md

Russian-language version of the memory rules — detailed guidelines for fact categories, trust scoring thresholds, and memory hygiene practices.

MEMORY.md

Initial memory seed — confirms Holographic memory is active and describes its purpose for deep analysis and contradiction detection.

Architecture

Hermes Agent (Nous Research)
├── Polza AI (custom endpoint)
├── Telegram Bot Gateway
│   ├── Voice Mode (faster-whisper STT)
│   └── Text/voice message handling
└── Holographic Memory (SQLite + FTS5)
    ├── fact_store (structured facts)
    ├── trust scoring (helpful/unhelpful feedback)
    └── search (probe, reason, contradict)

Key Components

Polza AI Integration

Custom LLM provider via OpenAI-compatible API. Configured through ~/.hermes/config.yaml with API key in ~/.hermes/.env.

Telegram Bot

• BotFather setup for token generation
• User ID-based access control
• Voice mode with local Whisper STT (no API key required)
• Gateway management via hermes gateway

Holographic Memory

Long-term memory system using SQLite + FTS5:

• fact_store: structured facts with categories (user_pref, project, tool, general, system)
• trust scoring: helpful (+1) / unhelpful (-1) feedback loop
• maintenance: 7-day review cycle, 30-day deprecation for unused facts

Files

├── .hermes.md      # Memory rules (English)
├── RULES.md        # Memory rules (Russian)
└── MEMORY.md       # Initial memory seed

Tech Stack

• Hermes Agent (Nous Research)
• Polza AI API
• Telegram Bot API
• Whisper STT (faster-whisper)
• SQLite + FTS5
• Python/Node.js runtime