HM-Arch
zhhjluka/hm-arch
Offline-first Python SDK for layered, human-like agent memory with forgetting and consolidation
HM-Arch is an offline-first Python SDK that gives agents a layered memory system with retention decay and consolidation. It lists Hermes diagnostics among its agent integrations, next to Codex and Claude Code hook installers.
What HM-Arch does
HM-Arch is an offline-first Python SDK that gives agents a layered, human-inspired memory behind one HMArch facade. Its layers run from L0 to L6: sensory, working, episodic, semantic, archive, procedural and meta-memory. Storage is local SQLite with deterministic retrieval, so tests, demos and the default runtime need no API keys, network access or external services.
Search results are ranked by relevance, retention and layer priority. Memories decay over time, can be scheduled for review, and can be removed with an explicit forget() call or a safe deletion window, while a consolidate() step extracts semantics, merges duplicates, resolves conflicts and archives old entries. Optional backends include OpenAI, DeepSeek and ChromaDB. The package ships hook installers for Codex and Claude Code and Hermes diagnostics, and is distributed through PyPI, npm and GitHub Releases.
Key features
- Seven memory layers, L0 to L6, behind one HMArch facade
- Retention-aware ranking by relevance, retention and layer priority
- Decay, review scheduling, safe deletion windows and explicit forget()
- Automatic consolidation that merges duplicates and resolves conflicts
- Local SQLite storage with optional OpenAI, DeepSeek and ChromaDB backends
- Hook installers for Codex and Claude Code plus Hermes diagnostics
When to use it
- Giving a custom agent long-term memory that works with no external services
- Letting memories fade or be archived instead of growing without limit
- Adding persistent context to Codex or Claude Code sessions through packaged hooks
Who it is for: Developers who want a local, inspectable memory layer for agents and are comfortable calling a Python SDK.
How it fits with Hermes Agent
It is a general agent memory SDK that supports several agents; for Hermes Agent the README lists diagnostics rather than a full provider integration.
How to install HM-Arch
These commands are copied from the project's README. Check the repository for the latest steps before you run them.
pip install hm-arch
# or: pipx install hm-archRequirements: Python 3.10+ for wheel installs; no API keys or network access for the default runtime
FAQ
What is HM-Arch?
HM-Arch is an offline-first Python SDK for agent memory. It stores experiences in seven layers, ranks retrieval by relevance and retention, forgets safely and consolidates knowledge over time.
Does HM-Arch work with Hermes Agent?
Partly. The README lists Hermes diagnostics among its agent-ready integrations, alongside packaged hook installers for Codex and Claude Code.
How do I install HM-Arch?
Run pip install hm-arch, or pipx install hm-arch. It is also published on npm as @hm-arch/installer and on GitHub Releases.
Similar memory for Hermes Agent
All memoryAgent-run installer that adds a portable, files-first memory workspace that survives fresh chats
atukunare Wiki Knowledge AgentAgent skill that verifies, translates and saves pasted links as wiki notes, with alerts for urgent items
spfcraze MindBankGraph-structured persistent memory for Hermes Agent and Claude Code with hybrid search and an MCP server
cx2002302-lang Zettelkasten Second MemoryPlugin that turns AI conversations into a linked Zettelkasten knowledge base with MCP tools
okjpg Agent Context KitPrivate local-first context retrieval for Hermes agents through a read-only MCP server
MaiHHConnect CausaMemLong-term causal memory system for AI agents with eight cognitive layers and intuition injection
Related guides: SOUL.md for Hermes Agent: what it is and how to write one · Run multiple Hermes agents with profiles