Hermes Memory-OS
btnalit/Hermes-Memory-OS
File-first memory and cognition runtime for long-running Hermes Agent profiles
Hermes Memory-OS is a file-first memory and cognition runtime for long-running Hermes Agent profiles. It gives an agent an inspectable memory lifecycle in which evidence is tracked, contradictions stay visible and durable beliefs must clear governance gates.
What Hermes Memory-OS does
Hermes stays the host and owns conversation, tools, scheduling, transport, retry and channel delivery. Memory-OS owns bounded memory state, retrieval, cognition artifacts, approval state machines and operational evidence. Experiences enter working memory, evidence accumulates, contradictions stay visible, and durable beliefs must pass governance gates, with every important transition auditable.
The README describes a layered stack: a memory kernel with canonical file storage and append-only lifecycle receipts, a recall engine combining FTS5, vector recall, memory and entity graphs, temporal recall and optional Hindsight recall, evidence and truth checks, governance and review queues, an expression gate with rate limits, and operational controls including a read-only dashboard. Cognition modules such as Inner Drive, Deep Reflection, Imagination Loop and Wandering Mind can stay disabled, shadowed, report-only or no-send until evidence and owner gates allow more. A governed right-brain module can associate freely, stay silent or share a thought without rewriting beliefs.
Key features
- Canonical file store with working, provisional, crystallized and permanent memory
- Recall through FTS5, vector search, memory and entity graphs and temporal lookup
- Evidence and truth checks, including contradiction lanes and confabulation detection
- Owner governance with approval state machines and audit receipts
- Optional cognition modules: Inner Drive, Deep Reflection, Imagination Loop, Wandering Mind
- Read-only dashboard, health probes and phased deploy tooling
When to use it
- Giving a long-running Hermes profile beliefs that carry evidence and can be corrected by the owner
- Keeping contradictory memories visible instead of silently overwriting them
- Auditing how an experience became a durable memory
Who it is for: Operators of long-running Hermes Agent profiles who want inspectable, governed memory rather than plain retrieval.
How it fits with Hermes Agent
It is tagged as a Hermes Agent plugin and runs per profile, with Hermes remaining the host for conversation, tools and delivery.
FAQ
What is Hermes Memory-OS?
Hermes Memory-OS is a memory and cognition runtime for long-running Hermes Agent profiles. It manages a file-based memory lifecycle with evidence checks, owner governance and audit trails.
Does Hermes Memory-OS work with Hermes Agent?
Yes. It is built for Hermes Agent profiles, with Hermes handling conversation, tools, scheduling and delivery while Memory-OS manages memory state and retrieval.
Is Hermes Memory-OS free and open source?
Yes. The repository is released under the MIT license.
Similar memory for Hermes Agent
All memoryHermes plugin that routes a local question to the skill, script, runbook or document to open first
labyrinth-analytics LoreConvoLocal SQLite memory for AI sessions that carries context across Claude Code, Cursor, Codex and Hermes
aristoapp Membase Plugin for HermesPersistent long-term memory for Hermes Agent using hybrid vector search and a knowledge graph
Mugen0815 hermes-cortexHybrid-search vault memory plugin for Hermes Agent, built on Markdown and Obsidian-style notes
MukundaKatta hermes-agentmemoryPull-model episodic memory plugin for Hermes Agent with real deletes and an audit trace
dax8it Total RecallLocal-first continuity memory with a hash-chained ledger and an optional Hermes provider plugin
Related guides: SOUL.md for Hermes Agent: what it is and how to write one · Run multiple Hermes agents with profiles