Hermes Memory Wiki

What Hermes Memory Wiki does

Sources are treated as the evidence layer and wiki pages as the human-readable synthesis layer. The vault has folders for sources, entities, concepts, syntheses and reports, plus index.md, inbox.md, WIKI.md and an AGENTS.md of rules for agents. Machine-readable output goes to .hermes-wiki/dist as agent-digest.json and claims.jsonl. Agent-generated text sits inside managed comment markers, so notes written by humans are not overwritten.

Eight tools in a memory_wiki toolset cover status, init, ingest, search, get, apply, compile and lint, and a /wiki slash command mirrors them. Search uses QMD for semantic matching and falls back to a local keyword scan. Config under plugins.memory-wiki sets the vault path (default ~/.hermes/wiki/main), the QMD collection and auto-embed, backlink and dashboard rendering, and claim aging at 30 days and staleness at 90. An opt-in bridge can sync with Honcho in either direction. The lint step looks for contradictions, stale claims, missing evidence and orphan pages.

Key features

• Vault with sources, entities, concepts, syntheses and generated reports
• Eight memory_wiki tools plus a /wiki slash command
• Compiled agent-digest.json and claims.jsonl for machine reading
• Lint checks for contradictions, stale claims, missing evidence and orphan pages
• QMD semantic search with a local keyword fallback
• Opt-in bidirectional sync with Honcho

When to use it

• Building a durable wiki from documents and notes that Hermes can query
• Tracking claims over time and flagging ones that have gone stale
• Keeping agent-written summaries separate from hand-written notes in the same vault

Who it is for: Hermes Agent users who want a git-friendly markdown knowledge base with evidence tracking instead of an opaque memory store.

How it fits with Hermes Agent

It is a Hermes Agent plugin that adds a memory_wiki toolset and a /wiki command, installed in ~/.hermes/plugins/memory-wiki.

FAQ