它为 AI 代理提供免费、本地、持久的内存层——无网络、无代币消耗——同时与上游设计字节兼容，并增加了排名搜索、口音归一化和原版缺少的原生 Windows 锁定功能。
A portable, dependency-free independent reimplementation of
Victor Taelin's OptMem design, wired
into Hermes Agent as a
standalone MemoryProvider.
License note. The upstream
VictorTaelin/OptMemrepo currently ships without an explicit license (all rights reserved by default). This repository is an independent reimplementation of the published design and memory format — it does not copy the upstream source. It is released here under MIT (see LICENSE). If/once upstream adopts a license, this plugin will align with it.
LOG.txt + decay TREE/ — uses the same on-disk format
as the original memo tool (fixed-width 320/288-byte records, identical
cover/decay math), so logs produced here are byte-compatible and
interchangeable with Taelin's implementation.recall("cacula") finds the caçula memory.msvcrt advisory locks (with spin/backoff), no WSL.Published as a standalone plugin repo per Hermes
CONTRIBUTING.md: third-party memory providers no longer land underplugins/memory/in the core tree. Install via~/.hermes/plugins/or pip.
Hermes ships cloud-backed memory providers (Honcho, Mem0, Hindsight). They are powerful for cross-session user modeling, but they impose a network round-trip and — on reasoning tiers — an LLM call on every recall.
OptMem is the cheap, always-available local layer: permanent, searchable memory that survives restarts, costs zero tokens, and never leaves your machine. Hermes allows one external provider at a time, so the two are complementary — OptMem is the durable default; a cloud provider is selected when richer modeling is wanted.
| Dimension | OptMem (local) | Cloud providers (e.g. Honcho) |
|---|---|---|
| Startup dependency | None — is_available() is always True | Needs API key / base URL |
| Per-call latency | Local BM25 (sub-ms to ms) | Network, sometimes an LLM |
| Per-call co |
OptMem memory provider plugin for Hermes Agent (append-only, decay-compressed, BM25 search)
根据分类、Topic 和编程语言匹配的相似项目。
Harness Engineering is a methodology for improving coding agent outputs by carefully crafting the environment around them—providing curated context, tools, and executable constraints that encode an organization’s nonfunctional requirements and cumulative lessons.
A 28.9 million parameter language model runs on an $8 ESP32-S3 microcontroller entirely on-device, generating simple stories at about 9.5 tokens per second.
mimic captures traffic from any iOS or web app and automatically generates a Python client library that lets you call the app's API like a regular library.