Hermes Atlas
Memory & knowledge

hermes-agentmemory

MukundaKatta/hermes-agentmemory

Pull-model episodic memory plugin for Hermes Agent with real deletes and an audit trace

In short

hermes-agentmemory is a standalone Hermes Agent memory plugin built on the agentmemory library. It builds recall summaries on demand with Claude, deletes events completely, and logs every prefetch to an audit trace file.

What hermes-agentmemory does

hermes-agentmemory uses a pull-model design: there is no background consolidation, every write is synchronous, and a recall builds its summary on demand with Claude. Deletes are immediate and complete, with no tombstone or derived summary left behind. Each prefetch appends a record with the intent, event ids, summary and a drift snapshot to $HERMES_HOME/agentmemory/trace.jsonl, so you can audit what entered the prompt.

The README positions it against Hermes' built-in memory backends such as Mem0, Honcho and Hindsight, which consolidate in the background and can leave a summary intact after the source event is removed. Hermes discovers the plugin by scanning $HERMES_HOME/plugins/ for a MemoryProvider subclass, so no extra registration is needed. The agent gets agentmemory_recall, agentmemory_forget and agentmemory_drift tools. The README states the trade-off plainly: the first turn of a new session pays a cost of 200 ms to 2 s for the on-demand summary. The code is under 600 lines of Python.

Key features

  • Synchronous writes with no background consolidation
  • Real deletes through agentmemory_forget, with no tombstone left
  • Audit trace in trace.jsonl for every prefetch
  • agentmemory_recall and agentmemory_drift tools
  • Configurable model, top-k and summary token budget through environment variables
  • Under 600 lines of Python

When to use it

  • Making sure a deleted memory is really gone from the agent
  • Checking exactly which past events were injected into a prompt
  • Using a small, readable memory plugin as a base for your own

Who it is for: Hermes Agent users who want auditable memory with deletions they can trust and who have an Anthropic API key.

How it fits with Hermes Agent

It is a drop-in Hermes memory plugin, activated with hermes config set memory.provider agentmemory.

How to install hermes-agentmemory

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

git clone https://github.com/MukundaKatta/hermes-agentmemory \
  "${HERMES_HOME:-$HOME/.hermes}/plugins/agentmemory"
pip install anthropic
hermes config set memory.provider agentmemory

Requirements: An ANTHROPIC_API_KEY for the on-demand summarizer and the anthropic Python package

Note: It needs an Anthropic API key for the on-demand summarizer, and the first turn of each new session pays a summary delay of roughly 200 ms to 2 s.

FAQ

What is hermes-agentmemory?

hermes-agentmemory is a memory plugin for Hermes Agent that stores episodic memory with real deletes and an audit trace. It builds summaries on demand instead of in the background.

Does hermes-agentmemory work with Hermes Agent?

Yes. It is cloned into the Hermes user plugins directory and activated with hermes config set memory.provider agentmemory.

What do I need to run hermes-agentmemory?

You need the anthropic Python package and an ANTHROPIC_API_KEY, since the on-demand summarizer calls Claude. The default model is claude-sonnet-4-5.

Similar memory for Hermes Agent

All memory

Related guides: SOUL.md for Hermes Agent: what it is and how to write one · Run multiple Hermes agents with profiles