memoryprofilespersonalizationvoice

Memory, Profiles & Personalization

By Hermes Agent· · 5 min read

Set up persistent memory so Hermes remembers you across sessions, create isolated profiles for work vs personal use, and enable voice interaction.

TLDR: Hermes remembers you across sessions through its memory system (enable with hermes config set memory.memory_enabled true). Profiles create completely isolated Hermes instances (hermes profile create "work"). Voice mode turns on with /voice on. Together they make Hermes feel like a personal assistant, not a stateless chatbot.

Key Takeaways

  • Memory stores facts about you and your environment; skills store procedures
  • Profiles give you fully independent Hermes instances (work vs personal)
  • Voice mode: /voice on for two-way, /voice tts for agent-only
  • Memory providers: built-in (free, local SQLite), Honcho, Mem0
  • The curator (hermes curator) runs automatic skill lifecycle maintenance
  • hermes profile export backs up entire profiles as tar.gz

Memory System

Memory is what makes Hermes feel intelligent between sessions. With cross-session memory enabled, it remembers who you are, your preferences, your environment, and lessons learned.

Enabling Memory

hermes config set memory.memory_enabled true
hermes config set memory.user_profile_enabled true

These take effect on next session (/reset or restart).

What Gets Saved

The agent automatically saves to memory:

  • Your preferences — “User prefers pnpm over npm”
  • Environment facts — “Running on Arch Linux with Intel CPU”
  • Corrections — “Don’t use apt, use pacman”
  • Workflow preferences — “User likes concise responses”

Two Memory Stores

StoreWhat it holdsWho writes it
memoryEnvironment facts, project conventions, lessons learnedAgent writes automatically
userYour name, preferences, communication styleAgent writes when corrected

Memory Providers

# Built-in SQLite (default, free, local — data never leaves your machine)
hermes config set memory.provider "built-in"

# Honcho (cloud, syncs across devices)
hermes config set memory.provider "honcho"

# Mem0 (cloud, advanced memory management)
hermes config set memory.provider "mem0"

Memory vs Skills

A common point of confusion:

MemorySkills
Facts about you and your environmentReusable procedures
”User uses Arch Linux""How to install packages on Arch”
Written by the agent automaticallyCreated by the agent on request
Loaded every session automaticallyLoaded on demand
Compact, one-sentence entriesStructured documents with steps

Rule of thumb: If it’s a fact → memory. If it’s a process → skill. See the Skills System guide for a deeper dive.

Profiles

Profiles are fully isolated Hermes instances. Each profile has its own config, session history, skills directory, memory database, and API keys.

Creating Profiles

# List existing
hermes profile list

# Create from scratch
hermes profile create "work"

# Clone current config (recommended)
hermes profile create "personal" --clone

# Set as default
hermes profile use "work"

Using Profiles

hermes --profile work          # One-off
hermes profile use "personal"  # Make default
hermes                         # Uses "personal"

Managing Profiles

hermes profile show work       # See details
hermes profile rename work dev # Rename
hermes profile delete personal # Remove (irreversible)
hermes profile export work     # Backup as tar.gz
hermes profile import work.tar.gz  # Restore

Use Cases

  • Work profile — project configs, company API keys, professional memory
  • Personal profile — personal projects, different providers, casual tone
  • Client profiles — separate memory per client, isolated skills
  • Testing profile — experimental configs, different models

Voice Mode

Hermes supports two-way voice interaction — speak to it, and it speaks back.

Setup

Configure both STT (speech-to-text) and TTS (text-to-speech):

# STT: voice to text
hermes config set stt.enabled true
hermes config set stt.provider local   # or groq, openai, mistral

# TTS: text to voice
hermes config set tts.provider edge    # or elevenlabs, openai

For local STT:

pip install faster-whisper

Voice Commands

/voice on      # Two-way voice (you speak, agent speaks back)
/voice tts     # Agent always speaks, you type
/voice off     # Text only (default)

Telegram and Discord voice messages are auto-transcribed when STT is enabled.

The Curator: Automatic Skill Maintenance

The curator manages your skills automatically, part of the skills lifecycle system:

hermes curator status             # See all skills and their state
hermes curator pin deploy-skill   # Protect from auto-archiving
hermes curator run                # Run maintenance now

It marks unused skills as stale, archives them after 60 days, but never deletes anything. Pinned skills are excluded from the lifecycle.

FAQ

Q: Does memory sync between CLI and gateway? If using the built-in provider, memory is local to the machine. If using Honcho or Mem0, memory syncs across all interfaces.

Q: Can I export my memory? Memory is stored in the SQLite database at ~/.hermes/. Export an entire profile with its memory using hermes profile export work.

Q: Will Hermes remember me if I switch profiles? No — each profile has isolated memory. That’s by design. Switch profiles with hermes --profile name.

Q: How is memory different from a skill? Memory stores facts (“user prefers pnpm”). Skills store procedures (“how to deploy the app”). See the Skills guide for details.

Q: Can I use voice on Telegram? Yes — Telegram voice messages are auto-transcribed when STT is enabled. Configure it with hermes config set stt.provider local and install faster-whisper.

Next Steps