Hermes Memory Providers: A Complete Breakdown for New Users
Hermes Memory Providers: A Complete Breakdown for New Users
Hermes 内存提供程序：新手完全指南
Hermes has a lot of memory options. If you’re new, the choices can be overwhelming — built-in memory, 8 external providers, different costs, different architectures. This guide breaks it all down so you can make the right call for your setup. Hermes 提供了多种内存选项。如果你是新手，这些选择可能会让你感到不知所措——内置内存、8 种外部提供程序、不同的成本、不同的架构。本指南将为你详细拆解，帮助你为自己的配置做出正确的选择。
First: Built-In Memory (Always Active)
第一部分：内置内存（始终处于活动状态）
Before we talk providers, understand that built-in memory is always on. It doesn’t cost anything, requires no setup, and works out of the box. 在讨论提供程序之前，请先了解内置内存是始终开启的。它无需任何费用，无需设置，开箱即用。
Two files in ~/.hermes/memories/:
~/.hermes/memories/ 下的两个文件：
| File | Purpose | Char Limit |
|---|---|---|
| MEMORY.md | Agent’s notes — environment facts, project conventions, lessons learned | 2,200 chars (~800 tokens) |
| USER.md | User profile — your name, preferences, communication style | 1,375 chars (~500 tokens) |
| 文件 | 用途 | 字符限制 |
|---|---|---|
| MEMORY.md | 智能体笔记 — 环境事实、项目约定、经验教训 | 2,200 字符 (~800 tokens) |
| USER.md | 用户资料 — 你的姓名、偏好、沟通风格 | 1,375 字符 (~500 tokens) |
Both are injected into the system prompt at the start of every session. The agent manages them automatically — it saves preferences you correct, environment facts it discovers, and conventions it learns. 这两个文件在每次会话开始时都会被注入到系统提示词中。智能体会自动管理它们——它会保存你修正的偏好、发现的环境事实以及学到的约定。
Key details: 关键细节：
- Entries are separated by
§delimiters. - 条目由
§分隔符隔开。 - The header shows usage % (e.g.,
MEMORY [67% — 1,474/2,200 chars]). - 头部显示使用百分比（例如：
MEMORY [67% — 1,474/2,200 chars]）。 - Above 80% capacity, the agent should consolidate before adding.
- 当容量超过 80% 时，智能体应在添加新内容前进行整合。
- Duplicate entries are auto-rejected.
- 重复条目会被自动拒绝。
- Entries are scanned for injection/exfiltration patterns for security.
- 出于安全考虑，条目会被扫描是否存在注入/数据外泄模式。
- Changes persist to disk immediately but appear in the system prompt at the next session (frozen snapshot — preserves LLM prefix cache).
- 更改会立即保存到磁盘，但在下一次会话时才会出现在系统提示词中（冻结快照 — 以保留 LLM 前缀缓存）。
For most new users, built-in memory is enough. It handles preferences, project facts, and daily workflow notes. You don’t need an external provider for a personal assistant setup. 对于大多数新用户来说，内置内存已经足够了。它能处理偏好、项目事实和日常工作笔记。对于个人助理设置，你不需要外部提供程序。
But you’ll want one when: 但在以下情况下，你可能需要外部提供程序：
- You have multiple Hermes profiles that should share knowledge.
- 你有多个需要共享知识的 Hermes 配置文件。
- You want the agent to learn and synthesize across sessions automatically.
- 你希望智能体能自动跨会话学习和综合信息。
- You’re running long conversations that exceed context limits.
- 你正在进行超出上下文限制的长对话。
- You need structured knowledge retrieval (entities, relationships, not just text blobs).
- 你需要结构化的知识检索（实体、关系，而不仅仅是文本块）。
The 8 External Memory Providers
8 种外部内存提供程序
All external providers are installed via: 所有外部提供程序均通过以下方式安装：
hermes memory setup # interactive picker
hermes memory status # check what's active
hermes memory off # disableOr set manually in ~/.hermes/config.yaml:
或者在 ~/.hermes/config.yaml 中手动设置：
memory:
provider: hindsight # or any of the 8Important: Only one external provider can be active at a time. All of them layer on top of built-in memory — they don’t replace it. 重要提示： 同一时间只能激活一个外部提供程序。它们都是在内置内存的基础上叠加的，并不会取代它。
Quick Comparison
快速对比
| Provider | Storage | Cost | Unique Angle | Best For |
|---|---|---|---|---|
| Hindsight | Local/Cloud | Free (local) | Knowledge graph + reflect synthesis | Highest accuracy, privacy |
| Holographic | Local SQLite | Free | HRR algebra + trust scoring, zero deps | Air-gapped, zero-install |
| OpenViking | Self-hosted | Free (AGPL) | Tiered L0/L1/L2 loading, 80-90% token savings | Self-hosted teams, cost optimization |
| Mem0 | Cloud | Freemium | Server-side LLM extraction, dual memory scope | Fastest setup |
| Honcho | Cloud/Self | Paid/Free | Dialectic user modeling | Multi-agent, deep user understanding |
| ByteRover | Local/Cloud | Freemium | Knowledge tree in Markdown | Pre-compression knowledge capture |
| RetainDB | Cloud | Paid | Hybrid search: vector + BM25 + reranking | Production search quality |
| SuperMemory | Cloud | — | Web-focused memory with browser integration | Web research workflows |
| 提供程序 | 存储 | 成本 | 独特优势 | 适用场景 |
|---|---|---|---|---|
| Hindsight | 本地/云端 | 免费 (本地) | 知识图谱 + 反思综合 | 最高准确度、隐私 |
| Holographic | 本地 SQLite | 免费 | HRR 代数 + 信任评分，零依赖 | 离线环境、零安装 |
| OpenViking | 自托管 | 免费 (AGPL) | 分层 L0/L1/L2 加载，节省 80-90% token | 自托管团队、成本优化 |
| Mem0 | 云端 | 免费增值 | 服务端 LLM 提取，双重内存范围 | 最快设置 |
| Honcho | 云端/自托管 | 付费/免费 | 辩证用户建模 | 多智能体、深度用户理解 |
| ByteRover | 本地/云端 | 免费增值 | Markdown 知识树 | 预压缩知识捕获 |
| RetainDB | 云端 | 付费 | 混合搜索：向量 + BM25 + 重排序 | 生产级搜索质量 |
| SuperMemory | 云端 | — | 专注网页的内存，带浏览器集成 | 网络研究工作流 |
Benchmark Snapshot
基准测试快照
Only two providers have published LongMemEval scores: 目前只有两家提供程序发布了 LongMemEval 分数：
| Provider | Score | Model |
|---|---|---|
| Hindsight | 91.4% | Gemini-3 |
| Hindsight | 89.0% | Open-source 120B |
| Mem0 | 67.6% | GPT-4o (LongMemEval-S variant) |
| 提供程序 | 分数 | 模型 |
|---|---|---|
| Hindsight | 91.4% | Gemini-3 |
| Hindsight | 89.0% | 开源 120B |
| Mem0 | 67.6% | GPT-4o (LongMemEval-S 变体) |
Hindsight is the clear retrieval accuracy leader. Others haven’t published comparable benchmarks. Hindsight 在检索准确度方面处于绝对领先地位。其他提供程序尚未发布可比的基准测试数据。
(Note: Due to length constraints, the detailed “Provider Deep Dives” section for each of the 8 providers follows the same logic as above. If you need the full text for the remaining sections, please let me know!) (注：由于篇幅限制，8 个提供程序的详细“深度解析”部分遵循上述逻辑。如果你需要剩余部分的全文，请告诉我！)