Hermes + Memanto: Persistent Memory Agent

This package adds Memanto as a memory agent for the Hermes agent. Memanto gives Hermes typed long-term memory with confidence and provenance, semantic recall, and RAG-style answers, backed by the Moorcheh vector platform.

Unlike a passive "memory layer", every namespace in Memanto is a first-class agent (memanto agent create/activate), so this provider maps one Hermes identity to one Memanto agent.

What you get

• Auto-recall — relevant memories are injected before each turn.
• Turn capture — conversation turns are stored as event memories.
• Explicit tools — memanto_remember, memanto_recall, memanto_answer.
• Memory mirroring — Hermes' built-in memory writes are echoed into Memanto.
• Profile isolation — agent_id: hermes-{identity} scopes memory per Hermes profile.

Why a standalone plugin?

Hermes' built-in providers under plugins/memory/ are a closed set; new memory backends ship as standalone plugins that users install into ~/.hermes/plugins/. Hermes discovers a memory provider as a directory under $HERMES_HOME/plugins/<name>/ containing an __init__.py that exposes register(ctx) plus a plugin.yaml. This package ships exactly that, plus an installer that drops it into place.

Prerequisites

• Python 3.10+
• A running Hermes agent install
• A Moorcheh API key (free tier: 100K ops/month)

Install

# 1. Install this package (pulls in the memanto SDK)
pip install hermes-memanto

# 2. Drop the plugin into your Hermes home (~/.hermes/plugins/memanto/)
hermes-memanto-install

# 3. Configure your key + select the provider
export MOORCHEH_API_KEY=...          # https://console.moorcheh.ai/api-keys
hermes config set memory.provider memanto

hermes memory setup will also list memanto once the plugin is installed, and writes MOORCHEH_API_KEY into ~/.hermes/.env for you.

From a source checkout

git clone https://github.com/moorcheh-ai/memanto.git
cd memanto/integrations/hermes-agents
pip install -e .
hermes-memanto-install            # or: hermes-memanto-install --hermes-home /path/to/.hermes

The installer copies hermes_memanto/provider.py verbatim as the plugin's __init__.py, so the installed plugin is self-contained and only needs the memanto SDK at runtime (declared in its plugin.yaml).

Configuration

After install, settings live in $HERMES_HOME/memanto.json:

Tools exposed to Hermes

Memory types follow Memanto's taxonomy: fact, preference, goal, decision, artifact, learning, event, instruction, relationship, context, observation, commitment, error.

Development

pip install -e ".[dev]"
pytest          # provider unit tests (no network; SdkClient is faked)
ruff check .

How it relates to the other integrations

All three talk to the same Moorcheh-backed Memanto agents, so memory written by one is recallable from the others when they share an agent_id.

Support

• Memanto Documentation
• Moorcheh API Keys
• Hermes agent