Hermes Graphiti Plugin: memory tool for Hermes Agent

Graphiti temporal knowledge-graph memory provider for Hermes Agent, stored in Neo4j

What Hermes Graphiti Plugin does

The plugin extracts entities, facts and relationships from conversations, stores them with timestamps in a Neo4j graph database and retrieves relevant context for later turns. Before each turn, prefetch() searches Graphiti and injects matching facts and entities into the system prompt. sync_turn() saves conversations in the background without blocking. The agent also gets a graphiti_search tool for explicit queries and a graphiti_delete_entity tool for removing stale entries.

Three search strategies are configurable: fast (BM25 only), balanced (BM25 plus cosine similarity) and quality (cross-encoder reranking). Reasoning tags from DeepSeek and Qwen3 are stripped before saving, and max_episode_content_chars caps episode size. Settings live in graphiti.json in the Hermes home, secrets go in .env, and memory.provider is set to graphiti in config.yaml. It needs a Neo4j database (4.4+ or 5.x) and an OpenAI-compatible LLM endpoint. The README marks it beta and not actively maintained.

Key features

Automatic recall of relevant facts before each turn
Background persistence of conversations to a Neo4j graph
graphiti_search and graphiti_delete_entity tools
Fast, balanced and quality search strategies
Reasoning-block stripping and episode content truncation

When to use it

Giving Hermes a time-aware memory of people, projects and decisions
Experimenting with graph-based memory next to the built-in MEMORY.md
Running memory against a local or self-hosted LLM and Neo4j

Who it is for: Hermes Agent users who want to try graph-based long-term memory and can run a Neo4j database.

How it fits with Hermes Agent

Built for Hermes Agent as a memory provider plugin, activated with memory.provider set to graphiti.

How to install Hermes Graphiti Plugin

These commands are copied from the project's README. Check the repository for the latest steps before you run them.

hermes plugin install https://github.com/p1s4/hermes-graphiti-plugin.git

Requirements: Neo4j 4.4+ or 5.x, an OpenAI-compatible LLM API endpoint, a running Hermes Agent and a Python 3.10+ environment

Note: The README labels it beta and experimental, tested for two weeks in a personal Docker setup, and not actively maintained.

FAQ

What is Hermes Graphiti Plugin?

Hermes Graphiti Plugin is a Hermes Agent memory provider that stores conversation knowledge as a temporal graph in Neo4j using Graphiti Core, and recalls it automatically on later turns.

What do I need to run Hermes Graphiti Plugin?

You need a Neo4j database (version 4.4+ or 5.x) reachable from Hermes, an OpenAI-compatible LLM API endpoint, and Hermes Agent on a Python 3.10 or newer environment.

Is Hermes Graphiti Plugin actively maintained?

No. The README labels it beta and experimental, tested for two weeks in a personal Docker environment, and not actively maintained. It is provided as is.