HHermes Explained

Memory

Memory

Selected knowledge that survives beyond a single prompt and can influence future turns.

Importance 5/514 backlinks1 source refs

Memory

What it is

Selected knowledge that survives beyond a single prompt and can influence future turns.

In plain terms

Memory is Hermes choosing which few details from a conversation are worth carrying into the future, the way a person might jot one line in a diary after a long day instead of transcribing every word that was said. Only the facts someone explicitly asked it to remember stick around; everything else disappears once the session ends, unless it was captured at the time.

Why it matters

A session with no memory layer starts from zero every time — no recollection of a user’s ongoing projects, preferences, or prior decisions. Memory is the mechanism that lets specific facts outlive the prompt that created them and get pulled back into a future turn’s context. It’s deliberately selective rather than a full transcript store, which is exactly what separates it from raw session history.

How it works

Underneath sits a single memory tool that lets the model add, replace, or remove short entries in two on-disk files: MEMORY.md for the agent’s own working notes and USER.md for facts about the person it’s talking to (tools/memory_tool.py). Both files load once at session start and get frozen into the system prompt as a fixed snapshot, keeping the model’s cached context stable; a write made mid-session updates the file on disk immediately but only appears in the prompt itself starting with the next session. MemoryManager (agent/memory_manager.py) is a separate layer that plugs in at most one external provider a user has configured; it doesn’t run this store itself, but gets notified whenever the store is written to, so it can mirror the entry out to that provider.

A concrete example

Say a user tells Hermes, over Telegram, that they always want cost estimates in Australian dollars, not US. Hermes treats that as a preference worth keeping, so it saves a short note to the user profile. Two weeks later, in a completely different session, the same person asks for a price on a new server, and the answer comes back in AUD without being reminded. If the memory file is already near its size limit, trying to add more gets rejected with the current entries shown, so something stale can be trimmed before the new note is saved.

How it connects

This note belongs to the Memory family. Its closest neighboring concepts are: Memory Tools, Memory Manager, Agent Runtime, Personal OS.

Source evidence

  • vendor/hermes-agent/agent/memory_manager.py:505-515#MemoryManager.prefetch_all — Verified Source Map evidence for MemoryManager. MemoryManager calls prefetch on each registered provider in order, skips empty results, isolates individual provider failures, and merges the remaining text.

Workflows

  • Daily Priority Ritual
  • Slack Second Brain
  • Why Hermes Feels Different: Memory, Self-Improving Skills, and an Agent That Keeps Working
  • Inside Hermes: How the Agent Loop, Context, Memory, Gateways, and Cron Jobs Work Together
  • The Ultimate Beginner’s Guide to Hermes: From First Installation to an AI Operations Team
  • Six Hermes Use Cases for a More Intentional Personal Operating System
  • How to Use Hermes Better Than Most People: Architecture, Project Isolation, and Scalable Workflows
  • Hermes Agent for Beginners: A Step-by-Step Route from Empty Server to Useful Automation
  • From Zero to a Personal AI Assistant: Building Hermes on a VPS Without Creating a Mess
  • Six Cost-Aware Hermes Workflows That Go Beyond Chat
  • Ten Underused Hermes Features That Reveal Its Real Power
  • Building a Personal AI System with Hermes, Local Models, Automation, and a Shared Knowledge Base