Hermes Observational Memory — Development skill for Claude Code

Hermes Observational Memory

**Unmaintained legacy software — September 5, 2026.** Final plugin release: **1.5.2**. No feature, compatibility or security updates are promised. Prefer Hermes built-in memory. See [migration](LEGACY.md); the installation instructions below are retained for legacy users and forks.

Observational Memory as a standalone Hermes memory-provider plugin.

It gives Hermes access to the same local-first memory store used by Claude Code, Codex, OpenCode, Kimi Code CLI, Grok, Cowork, and other OM-connected agents. Hermes can load compact startup context, search prior observations and reflections, store explicit notes, and optionally write Hermes turns back into Observational Memory.

Why Observational Memory

Observational Memory is a local-first memory backend. Memory stays in readable markdown files, with optional OM Cluster sync for sharing trusted records across machines. That makes it a good fit when you want cross-agent continuity without turning memory into an opaque hosted service.

Compared with hosted memory providers, this plugin is more inspectable and easier to audit. Compared with graph or hierarchy-first providers, it is focused on stable session continuity: an observer writes durable observations, a reflector condenses them into compact startup memory, and Hermes can recall more detail on demand.

Requirements

• Hermes with user-installed memory provider discovery. This is present in Hermes v2026.4.16 and newer; current upstream Hermes still routes standalone memory providers through the same MemoryProvider ABC, memory.provider, and $HERMES_HOME/plugins/ discovery path.
• observational-memory>=0.10.0,<0.11.
• Optional: an initialized OM Cluster if you want Hermes to share memory across machines.

Install

Install the plugin from GitHub:

hermes plugins install intertwine/hermes-observational-memory --no-enable

Memory providers are `kind: exclusive` plugins. They are activated throug