plur-hermes

Persistent memory plugin for Hermes Agent. Your agent corrected on Monday remembers on Tuesday.

Install

pip install plur-hermes

That's it. The plugin is auto-discovered by Hermes on startup. No other install needed.

What happens

Once installed, PLUR runs invisibly in the background:

Every turn: relevant memories are injected into the agent's context (via pre_llm_call hook)
Every response: corrections and insights are captured automatically (via post_llm_call hook)
Every session: episodes are recorded to a timeline (via on_session_end hook)

The agent also gets 16 tools it can call explicitly:

Tool	What it does
`plur_learn`	Store a correction, preference, or pattern
`plur_recall`	Search memories by topic
`plur_inject`	Get relevant context for a task
`plur_list`	List all stored engrams
`plur_forget`	Retire outdated knowledge
`plur_feedback`	Rate a memory (trains what surfaces next time)
`plur_capture`	Record an episode
`plur_timeline`	Query past episodes
`plur_status`	Health check
`plur_sync`	Cross-device sync via git
`plur_packs_list`	List installed knowledge packs
`plur_packs_install`	Install a community knowledge pack
`plur_extract_meta`	Distill cross-domain principles from your memories
`plur_meta_engrams`	List extracted meta-engrams
`plur_meta_submit_analysis`	Continue multi-turn extraction
`plur_validate_meta`	Test a principle against a new domain

How it works

Knowledge is stored as engrams — small assertions that strengthen with use and decay when irrelevant. Storage is plain YAML on disk at ~/.plur/. Search is fully local (BM25 + embeddings). Zero API calls, zero cloud, works offline.

The plugin calls the PLUR CLI under the hood via subprocess. If the CLI isn't installed globally, it auto-resolves via npx on first use (cached after that).

What makes PLUR different from Hermes built-in memory

Hermes has MEMORY.md (2,200 chars) and session search (FTS5). PLUR adds:

Feedback-trained retrieval — rate memories, good ones surface more, bad ones fade
Forgetting — retire outdated knowledge instead of growing forever
Hybrid search — BM25 + local embeddings + Reciprocal Rank Fusion
Cross-device sync — git-based, works across machines
Meta-engram extraction — distill transferable principles across domains
Knowledge packs — share curated engrams between agents

PLUR sits alongside Hermes memory, not replacing it. Your MEMORY.md and USER.md continue to work as before.

Configuration

The plugin works with zero configuration. Optional env vars:

Variable	Default	Description
`PLUR_PATH`	`~/.plur`	Storage directory
`PLUR_INJECT_MODE`	`fast`	Set to `hybrid` for embedding-based injection (slower, more accurate)

Requirements

Hermes Agent v0.5.0+
Python 3.10+
Node.js 18+ (for CLI, auto-resolved via npx if not installed globally)

License

Apache-2.0

Download files

Download the file for your platform. If you're not sure which to choose, learn more about installing packages.

Source Distribution

plur_hermes-0.11.0.tar.gz (52.1 kB view details)

Uploaded Jul 9, 2026 Source

Built Distribution

If you're not sure about the file name format, learn more about wheel file names.

The dropdown lists show the available interpreters, ABIs, and platforms. Enable javascript to be able to filter the list of wheel files.

plur_hermes-0.11.0-py3-none-any.whl (27.1 kB view details)

Uploaded Jul 9, 2026 Python 3

File details

Details for the file plur_hermes-0.11.0.tar.gz.

File metadata

Download URL: plur_hermes-0.11.0.tar.gz
Upload date: Jul 9, 2026
Size: 52.1 kB
Tags: Source
Uploaded using Trusted Publishing? No
Uploaded via: twine/6.2.0 CPython/3.11.1

File hashes

Hashes for plur_hermes-0.11.0.tar.gz
Algorithm	Hash digest
SHA256	`99f9d637283e639fa989aed461cb488f5b894e9df83d4a9684761b9b3a967a0f`
MD5	`25ef1bbe4fb67e1b28468f0514abd30b`
BLAKE2b-256	`3cd87ecc62db11f4f29dd8eaf5dc24fd502d21fd7f5f44dd553346cfe65fcbdf`

See more details on using hashes here.

File details

Details for the file plur_hermes-0.11.0-py3-none-any.whl.

File metadata

Download URL: plur_hermes-0.11.0-py3-none-any.whl
Upload date: Jul 9, 2026
Size: 27.1 kB
Tags: Python 3
Uploaded using Trusted Publishing? No
Uploaded via: twine/6.2.0 CPython/3.11.1

File hashes

Hashes for plur_hermes-0.11.0-py3-none-any.whl
Algorithm	Hash digest
SHA256	`4539f5d481e07cd702120d2da9a8a4f61d6dfe3d68250d730cd33ad66e7c6fa1`
MD5	`62d3a99d5c8010b00da10998c84f1de4`
BLAKE2b-256	`3bc186d2f583ac81ddfc9ad2fd1e0d73d41c972de19a64caee9062bf785c7d94`