memex8: memory tool for Hermes Agent

Self-hosted memory system with per-type decay and intent-weighted search, plus a Hermes memory plugin

What memex8 does

memex8 treats memory the way people do. Untouched memories fade on a Weibull curve that depends on their type, so preferences age slowly while events and requests fade quickly, with a floor at 0.05 so nothing is fully deleted. Queries are classified by intent and re-weighted, so a question about last week favors recency and a question about preferences favors importance. Related memories connect during a nightly slumber, and scattered fragments consolidate into denser summaries. Vector storage uses TurboVec for 8x compression, with Qdrant holding payload metadata and a web UI graph for visualization.

The plugins/memex8 directory is a Hermes Agent memory provider. It recalls context before each turn, syncs conversation turns while skipping trivial replies, ingests a session summary at the end and mirrors writes from the built-in memory tool. A pre-compress checkpoint archives the full transcript before lossy context rewrites, and setting compression.checkpoint_required to true makes it fail closed.

Key features

Per-type Weibull decay for preferences, facts, events and requests
Query intent classification that re-weights recency and importance
Associations and consolidation during nightly slumber
Hermes plugin with auto-recall, auto-sync and session-end ingestion
Pre-compress checkpoint that archives the transcript before context rewrites
Web UI graph of stored memories

When to use it

Replacing Hermes built-in memory with a backend that forgets at realistic rates
Keeping a full transcript archive before Hermes compresses context
Browsing connections between stored memories in a graph

Who it is for: Hermes Agent users who run their own memory backend and want decay, association and consolidation.

How it fits with Hermes Agent

A standalone third-party Hermes memory provider plugin, deliberately kept outside the in-tree plugins/memory directory, installed through hermes plugins install and hermes memory setup.

How to install memex8

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

hermes plugins install memex8
hermes memory setup

Requirements: A running memex8 engine backed by Qdrant; hermes memory setup asks for its URL and API key

Note: It is a personal project shared without support commitments, and the one-command install applies once the plugin is in the Hermes catalog.

FAQ

What is memex8?

memex8 is a self-hosted memory system for AI agents that models human memory, with per-type decay, intent-weighted queries and automatic associations. It includes a plugin for Hermes Agent.

How do I install memex8?

Run hermes plugins install memex8, then hermes memory setup and pick memex8, entering its URL and API key. The README says this one-command install applies once the plugin is in the Hermes catalog.

Is memex8 free and open source?

Yes. The repository is released under the MIT license, though the author describes it as a personal project with no support commitments.