TencentCloud/TencentDB-Agent-Memory
TencentDB Agent Memory is a Hermes Agent plugin providing a 4-tier progressive and local-first memory pipeline to enhance AI agents by reducing token usage and improving task success rates.
Awesome Hermes Plugins › Memory Providers
MukundaKatta/hermes-agentmemory
hermes-agentmemory is a standalone memory provider for Hermes Agent that wraps the author's agentmemory library. It is positioned against the consolidating backends Hermes ships with, such as Mem0, Honcho and Hindsight, which summarise in the background to make recall fast. The author argues that pattern costs two things: a delete is not always real, because removing an original event leaves any derived summary that already absorbed it intact, and memory injection is opaque, because background prefetch happens off the hot path so the user cannot see which past events entered the prompt until something goes wrong. This plugin inverts both. It does no background work, every write is synchronous, deletes are immediate and complete with no tombstone or leftover derived artifact, and every prefetch appends a trace record holding the intent, the event identifiers used, the generated summary and a drift snapshot to a JSONL file under the Hermes home directory, which the user can tail without touching the agent. Three tools are exposed to the model: `agentmemory_recall` returns the top matching past events along with the identifiers used, `agentmemory_forget` deletes by session or event identifier, and `agentmemory_drift` reports rolling-window retrieval quality for when recall starts feeling stale. Installation follows the official standalone path — Hermes stopped accepting new built-in memory providers — by cloning into the user-plugins directory that Hermes scans at startup, installing the summarizer's single dependency and setting `memory.provider`. Hermes discovers the provider by scanning plugin `__init__.py` files for a `MemoryProvider` subclass, so no registration step is needed. Environment variables configure the summarizer model, the number of events retrieved per prefetch, the summary token budget and the trace log path. The acknowledged trade-off is a 200ms to 2s cost on the first turn of a session because nothing is pre-warmed, in exchange for real deletes, an auditable trace and under 600 lines of readable Python. MIT licensed.
https://github.com/MukundaKatta/hermes-agentmemory
Also in Memory Providers
MemoriLabs/Memori
Memori is an agent-native memory infrastructure, offering an LLM-agnostic layer that converts agent execution and conversation into structured, persistent state for production systems, including a ...
basicmachines-co/basic-memory
Basic Memory is a local-first, knowledge graph-based memory system for AI agents, providing persistent, bidirectional knowledge access via Markdown files and the MCP protocol, including a Hermes pl...
agentscope-ai/ReMe
A local-first Markdown memory service with a Hermes Agent plugin that turns conversations and documents into linked, searchable notes using wikilinks, BM25 and optional embedding retrieval.
mnemosyne-oss/mnemosyne
Mnemosyne is a zero-dependency, sub-millisecond AI memory system for Hermes Agents and other LLM frameworks, focusing on efficient, local, and private knowledge retention.
ClaudioDrews/memory-os
Memory OS is a 7-layer memory operating system for Hermes Agent, providing persistent memory, structured facts, fabric recall, an auto-curated wiki, and surgical context injection, running locally ...
zhangfengcdt/memoir
Memoir is a high-performance semantic memory system for AI agents, offering Git-like version control for AI memory management to combat context contamination, token costs, and memory drift.
MaxFreedomPollard/Compartment
Offline, encrypted agentic memory vault that installs into Hermes Agent as a plugin and MCP server, keeping one transferable memory store shared by every agent on the machine.