Apache-2.0 本地优先、Markdown 原生的智能体记忆运行时；DeepSeek Harness 通过本地 HTTP API（/api/v2）或 MCP 桥接接入，获得持久长期记忆。
DSH 适配
兼容
作者声明
安全审计
未审计
最后核验
2026-08-21
许可证
Apache-2.0
01它能帮你完成什么？
Give DeepSeek Harness sessions persistent long-term memory
Durable memories stored as readable Markdown under ~/.everos, with local SQLite and LanceDB indexes and keyword search recall via the local HTTP API
Developers running DeepSeek Harness (dsh) or other coding agents who want cross-session context that survives restarts
Try the memory ingest -> extract -> index -> recall loop in minutes
A running local server with one-key setup: memory add, flush, Markdown persistence, cascade indexing, and keyword search
Builders evaluating agent memory options who want a no-infrastructure trial before committing
Ingest multimodal content into agent memory
Image, PDF, audio, and office-file ingestion through /api/v2/memory/add via the optional multimodal extra
Teams whose agent context includes documents, screenshots, and audio rather than plain text
02如何接入 DeepSeek Harness？
前置条件
- Python 3.12+
- One OpenRouter API key (https://openrouter.ai/keys)
安装步骤
- 01
Install the package: `uv pip install everos` (or `pip install everos`)
- 02
Run `everos init` — this creates `~/.everos/everos.toml` and `~/.everos/ome.toml`; replace the empty `api_key` in `~/.everos/everos.toml` with your OpenRouter key (use `everos init --root <path>` for a different memory root)
- 03
Start the server: `everos server start`
- 04
Add memories and search them via the HTTP API, e.g. POST /api/v2/memory/add, /api/v2/memory/flush, /api/v2/memory/search (use `"method": "keyword"` in the one-key setup)
- 05
Optional: `uv pip install 'everos[multimodal]'` for image / PDF / audio / office-file ingestion
验证接入成功
- Open a second terminal and run `curl http://127.0.0.1:8000/health` — look for `"status":"ok"`; with the one-key setup `capabilities.llm` is `true` while embedding and rerank remain `false`
- Add a tiny conversation via /api/v2/memory/add, flush it, then search it back via /api/v2/memory/search — you should see the stored memory in the response
03DSH 适配与能力边界
Local-first memory server that DeepSeek Harness can call through the local HTTP API (/api/v2/memory/add, /flush, /search) or community MCP bridges; the README documents EverOS as a universal long-term memory layer for AI coding assistants
Markdown-native memory persistence
conversations, files, and agent trajectories→canonical readable/editable/diffable .md files under the memory root (~/.everos), Git-versionable
writes Markdown files and syncs local SQLite and LanceDB indexes on diskcascade watcher re-syncs indexes when you edit .md files directlyLocal HTTP memory API with orthogonal retrieval
memory add / flush / search requests with user_id, agent_id, app_id, project_id, session_id→keyword (Tier 1) or hybrid search results from the local index; business endpoints live under /api/v2
starts a local server on 127.0.0.1:8000sends prompts to the configured OpenRouter endpoint when the llm capability is enabledTiered capability upgrades (embedding / rerank / multimodal)
optional [embedding], [rerank], [multimodal] sections in everos.toml→hybrid search, reflection and skill extraction; agentic search and Knowledge Wiki; image / PDF / audio / office ingestion
extra providers mean additional API calls and costsmultimodal extra pulls in everalgo-parser; office documents additionally require LibreOffice as a system dependencyOffline reflection (self-evolving memory)
accumulated episode clusters between sessions→merged episodes and refined user profiles and agent skills without online traffic
requires an embedding provider to be configured; mutates stored memory state during consolidation
04适合谁？何时不该用？
适合
- Developers running DeepSeek Harness (dsh) or other coding agents who want cross-session context that survives restarts
- Builders evaluating agent memory options who want a no-infrastructure trial before committing
- Teams whose agent context includes documents, screenshots, and audio rather than plain text
不适合
- The one-key OpenRouter setup is Tier 1 only (memory add/flush, Markdown persistence, cascade indexing, keyword search). Hybrid search defaults require an embedding provider, and unconfigured features return HTTP 422.
05兼容性、维护与安全提示
- The one-key OpenRouter setup is Tier 1 only (memory add/flush, Markdown persistence, cascade indexing, keyword search). Hybrid search defaults require an embedding provider, and unconfigured features return HTTP 422.
- Office document ingestion (.doc/.docx/.ppt/.pptx/.xls/.xlsx) requires LibreOffice installed on the host; without it, office uploads return HTTP 415 (PDF/image/audio/HTML/email parsing is unaffected).
- Core flows require Python 3.12+ and a paid OpenRouter API key stored in ~/.everos/everos.toml; the README does not document a fully offline (key-less) production mode beyond the educational demo.
Apache-2.0 · actively maintained (latest release v1.2.3, 2026-08-07)
06常见问题
DeepSeek Harness 如何接入 EverOS？
把 EverOS 作为本地记忆服务运行（everos server start），在 dsh 工作流中调用其 HTTP API：/api/v2/memory/add、/api/v2/memory/flush、/api/v2/memory/search。README 将 EverOS 定位为 AI 编码助手的通用长期记忆层，用例区还有多个编码 CLI 集成示例可参考。
安装前需要准备什么？
Python 3.12+ 和一个 OpenRouter API 密钥。如果只想先体验摄入→提取→索引→召回的完整流程，可直接运行 everos demo，无需任何密钥。
记忆数据存在哪里？
全部存在本地 ~/.everos 目录（也可用 everos init --root <path> 指定其他路径），以可读、可编辑的 Markdown 文件持久化，并用本地 SQLite 和 LanceDB 建索引，不需要 MongoDB、Elasticsearch 或 Redis。
单密钥（Tier 1）模式有什么限制？
只覆盖服务启动、记忆写入与 flush、Markdown 持久化、级联索引和关键词检索；混合检索、反思和知识 Wiki 需要额外配置 embedding/rerank 提供商。缺失能力会在 /health 中报告，未配置的功能调用会返回 HTTP 422。
能摄入文档和图片吗？
可以。安装可选的 everos[multimodal] 扩展后支持图片、PDF、音频、HTML、邮件和 Office 文档；其中 Office 格式还要求主机安装 LibreOffice，否则会返回 HTTP 415。
07相关的 DSH 工作流
ruflo
作者 ruvnet
原版智能体元框架（meta-harness）：支持多智能体集群部署、自主工作流编排与对话式 AI 系统构建，内置自适应记忆、自学习智能与 RAG 检索，原生集成 Claude Code / Codex / Hermes 等大量工具。
reactive-resume
作者 amruthpillai
注重隐私的开源简历生成器，安全、可定制、可移植，永久免费，支持自托管。
yao
作者 yaoapp
在桌面、移动端、浏览器或 API 上统一管理你的智能体与工作区，看板式任务追踪，支持自托管。
openpencil
作者 zseven-w
全球首个开源 AI 原生矢量设计工具，支持并发智能体团队，可在画布上直接通过提示词生成 UI。
08数据与来源
Universal long-term memory layer for AI coding assistants, powered by EverOS.
EverOS is a Python library and local-first memory runtime for agents and makers. It gives one portable memory layer acro…
页面基于项目公开文档、仓库元数据和 DSH Plugins 的结构化解析生成；最后核验于 2026-08-21。发现错误？提交更正。
最佳 DeepSeek Harness 插件
从全目录挑出的 12 个值得优先安装的插件，覆盖各个分类。